DETAILED ACTION
Applicant's amendment of January 29, 2026 overcomes the following:
Drawing objections
Specification objections related to title of the invention
Applicant has amended claims 1, 13 and 25. Claims 1-13, 25, 27-30, and 32-34 are pending.
Response to Arguments
Applicant’s arguments filed on January 29, 2026 with respect to pending claims have been considered but are moot in view of the new ground(s) of rejection. The amended claims resulted in changes to the scope and contents; therefore, the grounds of rejection are modified accordingly.
Specification
The specification is objected to as failing to provide proper antecedent basis for the claimed subject matter. See 37 CFR 1.75(d)(1) and MPEP § 608.01(o). Correction of the following is required:
Claim 1 now recites the limitation “wherein the MRI information comprises a single echo acquired without requiring a phase encoding gradient for spatial encoding” in lines 6-7 of the claim. The aforementioned claimed subject matter (a phase encoding gradient for spatial encoding) has no antecedent basis the specification.
Claim 13 now recites the limitation “wherein the MRI information comprises a single echo acquired without requiring a phase encoding gradient for spatial encoding” in lines 5-6 of the claim. The aforementioned claimed subject matter (a phase encoding gradient for spatial encoding) has no antecedent basis the specification.
Claim 25 now recites the limitation “wherein the MRI information comprises a single echo acquired without requiring a phase encoding gradient for spatial encoding” in lines 4-5 of the claim. The aforementioned claimed subject matter (a phase encoding gradient for spatial encoding) has no antecedent basis the specification.
Claim Rejections - 35 USC § 112
Claims 1, 13, and 25 are rejected under 35 U.S.C. 112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, as failing to comply with the written description requirement. The claim(s) contains subject matter which was not described in the specification in such a way as to reasonably convey to one skilled in the relevant art that the inventor or a joint inventor, or for applications subject to pre-AIA 35 U.S.C. 112, the inventor(s), at the time the application was filed, had possession of the claimed invention.
Claim 1 now recites the limitation “wherein the MRI information comprises a single echo acquired without requiring a phase encoding gradient for spatial encoding” in lines 6-7 of the claim. The aforementioned claimed subject matter has no antecedent basis the specification.
Par. [0029-42] of the specification of this application indicate “utilize reconstruction-only approach, referred to as single echo reconstruction (“SER”) to facilitate rapid 128×128 MR imaging using a 64-channel coil without phase encoding, for example, Tacq=Tencode*NSA, where Tencode can be the time to acquire one echo… exemplary system, method, and computer-accessible medium can facilitate significant reduction of RF power, PNS, and gradient noise… FIG. 1B can show an exemplary pypulseq (4,5) SER pulse sequence timing diagram with no phase encoding (Gy=0)… For example, the pypulseq coded sequence can acquire the central line in Cartesian k-space (phase encoding=0)… exemplary SER procedure(s) (i) may not require additional RF transmit channels for spatial encoding”, for example.
However, the examiner was not able to clearly ascertain where support was found in the original disclosure for the claimed “wherein the MRI information comprises a single echo acquired without requiring a phase encoding gradient for spatial encoding” recited in lines 6-7 of the claim.
Claim 13 now recites the limitation “wherein the MRI information comprises a single echo acquired without requiring a phase encoding gradient for spatial encoding” in lines 5-6 of the claim. The aforementioned claimed subject matter has no antecedent basis the specification.
Par. [0029-42] of the specification of this application indicate “utilize reconstruction-only approach, referred to as single echo reconstruction (“SER”) to facilitate rapid 128×128 MR imaging using a 64-channel coil without phase encoding, for example, Tacq=Tencode*NSA, where Tencode can be the time to acquire one echo… exemplary system, method, and computer-accessible medium can facilitate significant reduction of RF power, PNS, and gradient noise… FIG. 1B can show an exemplary pypulseq (4,5) SER pulse sequence timing diagram with no phase encoding (Gy=0)… For example, the pypulseq coded sequence can acquire the central line in Cartesian k-space (phase encoding=0)… exemplary SER procedure(s) (i) may not require additional RF transmit channels for spatial encoding”, for example.
However, the examiner was not able to clearly ascertain where support was found in the original disclosure for the claimed “wherein the MRI information comprises a single echo acquired without requiring a phase encoding gradient for spatial encoding” recited in lines 5-6 of the claim.
Claim 25 now recites the limitation “wherein the MRI information comprises a single echo acquired without requiring a phase encoding gradient for spatial encoding” in lines 4-5 of the claim. The aforementioned claimed subject matter has no antecedent basis the specification.
Par. [0029-42] of the specification of this application indicate “utilize reconstruction-only approach, referred to as single echo reconstruction (“SER”) to facilitate rapid 128×128 MR imaging using a 64-channel coil without phase encoding, for example, Tacq=Tencode*NSA, where Tencode can be the time to acquire one echo… exemplary system, method, and computer-accessible medium can facilitate significant reduction of RF power, PNS, and gradient noise… FIG. 1B can show an exemplary pypulseq (4,5) SER pulse sequence timing diagram with no phase encoding (Gy=0)… For example, the pypulseq coded sequence can acquire the central line in Cartesian k-space (phase encoding=0)… exemplary SER procedure(s) (i) may not require additional RF transmit channels for spatial encoding”, for example.
However, the examiner was not able to clearly ascertain where support was found in the original disclosure for the claimed “wherein the MRI information comprises a single echo acquired without requiring a phase encoding gradient for spatial encoding” recited in lines 4-5 of the claim.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 13, 25, and 30 are rejected under 35 U.S.C. 103 as being unpatentable over Kyriakos et al. (US Patent Publication No. 6680610 B1), hereafter referred to as Kyriakos, in view of Levin et al. (US PG Publication No. 2014/0266195 A1), hereafter referred to as Levin, in further view of James et al. (US PG Publication No. 2017/0261584 A1), hereafter referred to as James.
Regarding claim 13, Kyriakos discloses a system for reconstructing at least one portion of at least one image of at least one patient (Col. 1: apparatus and methods for magnetic resonance imaging, also known as magnetic resonance imaging (MRI)… wherein magnetic resonance data is acquired in parallel using an array of receiver coils at least partially Surrounding the object of interest; Col. 6: An object or subject (not shown) may be placed in the imaging volume within the cylinder 10 for examination using NMR phenomena. The subject is placed on the z-axis or the line 8, and is located within the coil 12. The coil 12 is representative of devices used to generate RF fields in the subject placed in the system for examination), comprising:
a computer hardware arrangement configured (Col. 7: system also includes a central processor 20 which can be a central processing unit (CPU), a memory device 30, and a display device 40… FIG. 2 is a flow diagram illustratively showing the basic sequence of the method of the invention. Thus, it will be seen that the present method contemplates that an apparatus such as that described above will be provided. In this apparatus, the CPU is preprogrammed to carry out numerous operations) to:
receive magnetic resonance imaging (MRI) information for the at least one patient (Col. 1: apparatus and methods for magnetic resonance imaging, also known as magnetic resonance imaging (MRI)… wherein magnetic resonance data is acquired in parallel using an array of receiver coils at least partially Surrounding the object of interest; Col. 2: method generally includes the following steps. First, a magnetic resonance imaging device is provided. This device typically includes a magnet system providing a background magnetic field in an imaging volume, a central processor, a memory device, a RF coil surrounding the imaging volume, and a plurality of radio frequency receiver coils defining a multi-dimensional array thereof disposed about the imaging volume. A two-dimensional sensitivity profile for each receiver coil in the multi-dimensional array is then computed and recorded to the memory device. Thereafter, a plurality of magnetic resonance signals of the object of interest located within the imaging volume is acquired from each receiver coil and recorded to the memory device. An image of the object of interest in a desired plane extending transversely through the imaging volume of the magnetic resonance imaging device then is reconstructed line-by line by the central processor. This reconstruction combines the inverse of the matrix of the sensitivity profiles of each receiver coil and the matrix of the recorded MR data signals together to provide an image that may be displayed or printed; Col. 4: the apparatus preferably constitutes a magnetic resonance imaging device including an imaging volume, preprogrammed central processor, memory device, a body coil surrounding said imaging volume, a plurality of radio frequency receiver coils defining a multi-dimensional array thereof disposed about said imaging volume, and a display device… Each receiver coil in the multi-dimensional array has an associated two-dimensional sensitivity profile… the central processor is adapted to reconstruct, and to display on the display device, line-by-line, a two-dimensional image taken in a selected plane extending transversely through the object of interest; Col. 6: present invention provides a magnetic resonance imaging system. The system includes a plurality of gradient coils that produce spatially encoded gradients imposed upon a background magnetic field Bo within the volume in which the object to be examined is placed… object or subject (not shown) may be placed in the imaging volume within the cylinder 10 for examination using NMR phenomena. The subject is placed on the z-axis or the line 8, and is located within the coil 12. The coil 12 is representative of devices used to generate RF fields in the subject placed in the system for examination; Col. 7: In MRI systems, it is necessary to provide various gradient coils (not shown) for producing spatially encoding gradients that are imposed upon the static magnetic field within the region in which the subject to be examined is placed also are provided… the system also includes a central processor 20 which can be a central processing unit (CPU), a memory device 30, and a display device 40; receive magnetic resonance imaging (MRI) information for the at least one patient (e.g. apparatus (i.e. a system) and methods for magnetic resonance imaging (MRI) include reconstructing an image of an object or subject of interest to be examined, such as a patient (i.e. reconstructing at least one image of at least one patient), for example, by using sensitivity profiles of an array of RF receiver coils at least partially (i.e. at least one portion) surrounding the object or subject of interest (i.e. reconstructing at least one portion of at least one image of at least one patient), for example, by using a programmed central processor (i.e. a computer hardware arrangement), which performs the following operation steps (i.e. comprising: a computer hardware arrangement configured to), including acquiring (i.e. receiving, obtaining, etc.) a plurality of magnetic resonance signals of the object or subject of interest (i.e. receive magnetic resonance imaging (MRI) information for the at least one patient), as indicated above), for example);
generate a plurality of coil sensitivity weighted projections based on the MRI information;
invert a column in the plurality of coil sensitivity weighted projections to generate inverted column information; and
reconstruct the at least one portion of the at least one image based on the inverted column information (Col. 2-3: parallel encoding in MRI is achieved by using the sensitivity profiles of an array of RF receiver coils at least partially surrounding the object of interest… Wi(x,y) represents the 2D sensitivity profile of the ith coil of the array, and ρ(x, y) is an image slice in a selected (x, y) plane… equation represents a projection of the phase modulated image ρ(x,y) onto the x-axis. Further, this signal can be represented in discrete form… apparatus and method of the present invention use phase modulated projections of the received magnetic resonance data onto the frequency encoded (x-) axis, weighted by the 2D sensitivity profiles of the coils in the array, in order to reconstruct ρ(x,y) column by column (i.e., orthogonal to the x-axis)… A two-dimensional sensitivity profile for each receiver coil in the multi-dimensional array is then computed and recorded to the memory device. Thereafter, a plurality of magnetic resonance signals of the object of interest located within the imaging volume is acquired from each receiver coil and recorded to the memory device. An image of the object of interest in a desired plane extending transversely through the imaging volume of the magnetic resonance imaging device then is reconstructed line-by line by the central processor. This reconstruction combines the inverse of the matrix of the sensitivity profiles of each receiver coil and the matrix of the recorded MR data signals together to provide an image that may be displayed or printed; Col. 4: an object of interest is located in the imaging volume of the magnetic resonance imaging device. Magnetic resonance data from the plane of interest is acquired from each of the RF receiver coils and stored in memory… determining and storing in memory the point-by-point time difference of signal reception by each of the receiver coils. This information may be combined with the sensitivity information in the inverse of the matrix of the sensitivity profile so as to provide a phase compensated, load-weighted, point-by-point, inverse, sensitivity profile matrix for each said receiver coil… The receiver coils are adapted to acquire simultaneously a plurality of magnetic resonance signals from an object of interest located within the imaging volume, and to transfer the same to the memory device… the central processor is adapted to reconstruct, and to display on the display device, line-by-line, a two-dimensional image taken in a selected plane extending transversely through the object of interest…This is accomplished by multiplying the inverse of the matrix of sensitivity profiles of the receiver coils and the matrix of data signals acquired by the receiver coils, and displaying the resultant product on the display device; Col. 8-9: sensitivity profile information from a number of receiver coils is used in order to minimize the number of acquisitions needed to estimate and reconstruct ρ(x, y). Taking the Fourier Transform of equation (1) along the x direction when a phase encoding gradient Gg y is applied yields… the phase-modulated projection of the sensitivity-weighted image onto the x-axis… With reference to FIG. 4, the Fourier Transform of the signals received in all three coils respectively is equal to the projection of the phase encoded image weighted by the three profiles W1(i,j), W2(i,j), W3(i,j) respectively onto the x axis… Assume the image to be of size 3×3 as shown in FIG. 4, where each pixel has an intensity value denoted by r(i,j), where i is the column number and j is the row number… in order to reconstruct the image [p(m,n)], a matrix pseudoinverse of [Amj(k,y)] is computed for every position mj along the x-axis. This yields a column by column reconstruction; Col. 10-11: Gx and Gy represent the gradients applied in the x and y directions respectively. Then taking the fourier transform with respect to x provides… the phase modulated projection of the image r (x,y) weighted by the sensitivity profile Wk(i,j) onto the x-axis… In order to appropriately account for all of the frequencies included in the image, the choice of the phase modulations used in the inversion matrix should be determined by the frequency content of the sensitivity profile; generate a plurality of coil sensitivity weighted projections based on the MRI information;
invert a column in the plurality of coil sensitivity weighted projections to generate inverted column information; and
reconstruct the at least one portion of the at least one image based on the inverted column information (e.g. apparatus and methods for magnetic resonance imaging (MRI) include reconstructing an image of an object or subject of interest to be examined, such as a patient (i.e. the at least one image), for example, by using sensitivity profiles of an array of RF receiver coils at least partially (i.e. at least one portion) surrounding the object or subject of interest (i.e. reconstruct the at least one portion of the at least one image), for example, by computing a two-dimensional sensitivity profile for each receiver coil in a multi-dimensional array (i.e. matrix) that is recorded to the memory device, by using phase modulated projections of received magnetic resonance data onto a frequency encoded x-axis, weighted by 2D sensitivity profiles of the coils in the array (i.e. generate a plurality of coil sensitivity weighted projections based on the MRI information), for example, and using a matrix pseudoinverse computed for every position along the x-axis (i.e. invert a column in the plurality of coil sensitivity weighted projections to generate inverted column information), which yields a column by column reconstruction (i.e. reconstruct the at least one portion of the at least one image based on column information), including an image of the object or subject of interest reconstructed by the programmed central processor, for example, by combining the inverse of the matrix of the sensitivity profiles of each receiver coil and the matrix of the recorded MR data signals together to provide an image that is displayed or printed, as indicated above), for example), but fails to teach the following as further recited in claim 13.
However, Levin teaches wherein the MRI information comprises a single echo acquired without requiring a phase encoding ( Par. [0035-41]: method for magnetic resonance imaging is proposed, which is based on transverse magnetization excitation in an additional (reference) slice location and a motion compensation based on a determined reference echo signal phase shift… the at least one reference echo signal is formed without phase encoding; Par. [0053-68]: acquiring the at least one first echo signal with phase encoding and the at least one reference echo signal without phase encoding… a processor means in order to compensate a motion-related phase shift of the at least one echo signal based on the at least one reference phase shift… receiver means is additionally adjusted to acquire the at least one reference echo signal without phase encoding… present invention may use a phase shift of the non-phase encoded reference echo-signal accumulated due to applying the diffusion-weighting gradients in order to characterize bulk motion and tissue deformation and to compensate their effect for adjusting the magnetic resonance image; wherein the MRI information comprises a single echo acquired without requiring a phase encoding (e.g. method for magnetic resonance imaging (MRI) is proposed, in which at least one reference echo signal (i.e. a single echo) is formed (i.e. acquired, obtained, constructed, etc.) without phase encoding (i.e. wherein the MRI information comprises a single echo acquired without requiring a phase encoding), as indicated above), for example).
Kyriakos and Levin are considered to be analogous art because they pertain to MR image processing applications. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to modify the apparatus and methods for magnetic resonance imaging including reconstructing of an image of an object or subject of interest to be examined (as disclosed by Kyriakos) with wherein the MRI information comprises a single echo acquired without requiring a phase encoding (as taught by Levin, Abstract, Par. [0035-41, 53-68]) to characterize bulk motion and tissue deformation and to compensate their effect for adjusting the magnetic resonance image (Levin, Abstract, Par. [0035-41, 53-68]).
The combination of Kyriakos and Levin, as a whole, teaches the system, as indicated above, but fails to teach the following as further recited in claim 13.
However, James teaches without requiring a phase encoding gradient for spatial encoding (Par. [0020-21]: Non-invasive techniques that do not rely on use of ionizing radiation or radioactive tracers allow the most leeway for early diagnosis and repeat measurement to monitor disease progression and response to therapy. Magnetic Resonance Imaging (MRI), which provides tunable tissue contrast, is just such a non-invasive technique… embodiments disclosed herein provide a method for pathology assessment employing tissue texture using magnetic resonance (MR) which may be used integrally with an MR imaging technique; Par. [0065-94]: MRI Magnetic Resonance Imaging… MRS Magnetic Resonance Spectroscopy… NMR Nuclear Magnetic Resonance… Several benefits result from acquiring data after the gradient is switched off for single-k-value sampling in a reduced volume (the VOI). By proper pulse sequencing, the echo record window can be designed such that recording begins with the highest k-values of interest, as signal level is highest at the echo peak… In certain embodiments disclosed herein, the approach to k-space filling is to acquire only the set of k-values needed for texture evaluation in the targeted pathology, with data acquired after the gradient is switched off; Par. [0116-122]: Within a given acquisition in standard MR practice, there are M samples which are acquired of the echo… If the entire echo is used to measure one k-value, the receive bandwidth can be adjusted so as to pass the most abundant resonant peaks in the underlying NMR spectrum, and attenuate frequencies above them… Taking a straight MRS spectrum (no structural phase encodes), would yield a spectrum consisting primarily of peaks… the full NMR spectrum may be extracted (without any phase encoding gradients… isochromats of interest can be extracted by acquiring N samples of the echo… Since the echo is being played out with no gradient, the strength of the resulting signal at the Isochromat of interest will correspond to the (complex-valued) k-value coefficient of interest; Par. [0285]: methods disclosed herein entail acquisition within either a single VOI or within multiple, interleaved VOIs, within one TR. Data is acquired without use of a spatial encoding gradient to form an image; without requiring a phase encoding gradient for spatial encoding (e.g. method for pathology assessment employing tissue texture using magnetic resonance (MR) used integrally with an MR imaging (i.e. MRI) technique, for example, includes acquiring data after gradient is switched off for single-k-value sampling in a reduced volume (VOI), for example, and by proper pulse sequencing, an echo record window is designed such that recording begins with highest k-values of interest, as signal level is highest at the echo peak (i.e. a single echo acquired), for example, in which a full Nuclear Magnetic Resonance (NMR) spectrum is extracted (i.e. acquired, obtained, etc.) without any phase encoding gradients (i.e. without requiring a phase encoding gradient), for example, including data acquired without use of a spatial encoding gradient to form an image (i.e. without requiring a phase encoding gradient for spatial encoding), as indicated above), for example).
Kyriakos, Levin, and James are considered to be analogous art because they pertain to MR image processing applications. Therefore, the combined teachings of Kyriakos, Levin, and James, as a whole, would have rendered obvious the invention recited in claim 13 with a reasonable expectation of success in order to modify the apparatus and methods for magnetic resonance imaging including reconstructing of an image of an object or subject of interest to be examined (as disclosed by Kyriakos) with without requiring a phase encoding gradient for spatial encoding (as taught by James, Abstract, Par. [0020-21, 65-94, 116-122, 285]) to circumvent the problem of signal degradation due to patient motion, to provide a method for pathology assessment employing tissue texture using magnetic resonance (MR) which may be used integrally with an MR imaging technique, by acquiring MRI data without use of a spatial encoding gradient to form an image which significantly shortens acquisition time and, combined with a narrowly targeted acquisition in k-space, enables acquisition of requisite data fast enough to provide immunity to subject motion (James, Abstract, Par. [0003-6, 11, 20-21, 65-94, 116-122, 285]).
Regarding claim 25, is a corresponding method claim rejected as applied to the apparatus claim 13 above.
Regarding claim 30, claim 25 is incorporated and the combination of Kyriakos, Levin, and James, as a whole, teaches the system (Kyriakos, Col. 1), at least one of: wherein the MRI information includes a signal collected over a time t and channels q, wherein the inverted column information includes line-intensity profiles, or further comprising:
(a) inverting a further column in the plurality of coil sensitivity weighted projections to generate further inverted column information;
(b) reconstructing at least one further portion of the at least one image based on the further inverted column information; and
(c) repeating procedures (a) and (b) until the at least one image is reconstructed in its entirety (Kyriakos, Col. 2-3: parallel encoding in MRI is achieved by using the sensitivity profiles of an array of RF receiver coils at least partially surrounding the object of interest… Wi(x,y) represents the 2D sensitivity profile of the ith coil of the array, and ρ(x, y) is an image slice in a selected (x, y) plane… equation represents a projection of the phase modulated image ρ(x,y) onto the x-axis. Further, this signal can be represented in discrete form… apparatus and method of the present invention use phase modulated projections of the received magnetic resonance data onto the frequency encoded (x-) axis, weighted by the 2D sensitivity profiles of the coils in the array, in order to reconstruct ρ(x,y) column by column (i.e., orthogonal to the x-axis)… A two-dimensional sensitivity profile for each receiver coil in the multi-dimensional array is then computed and recorded to the memory device. Thereafter, a plurality of magnetic resonance signals of the object of interest located within the imaging volume is acquired from each receiver coil and recorded to the memory device. An image of the object of interest in a desired plane extending transversely through the imaging volume of the magnetic resonance imaging device then is reconstructed line-by line by the central processor. This reconstruction combines the inverse of the matrix of the sensitivity profiles of each receiver coil and the matrix of the recorded MR data signals together to provide an image that may be displayed or printed; Col. 4: an object of interest is located in the imaging volume of the magnetic resonance imaging device. Magnetic resonance data from the plane of interest is acquired from each of the RF receiver coils and stored in memory… determining and storing in memory the point-by-point time difference of signal reception by each of the receiver coils. This information may be combined with the sensitivity information in the inverse of the matrix of the sensitivity profile so as to provide a phase compensated, load-weighted, point-by-point, inverse, sensitivity profile matrix for each said receiver coil… The receiver coils are adapted to acquire simultaneously a plurality of magnetic resonance signals from an object of interest located within the imaging volume, and to transfer the same to the memory device… the central processor is adapted to reconstruct, and to display on the display device, line-by-line, a two-dimensional image taken in a selected plane extending transversely through the object of interest…This is accomplished by multiplying the inverse of the matrix of sensitivity profiles of the receiver coils and the matrix of data signals acquired by the receiver coils, and displaying the resultant product on the display device; Col. 8-9: sensitivity profile information from a number of receiver coils is used in order to minimize the number of acquisitions needed to estimate and reconstruct ρ(x, y). Taking the Fourier Transform of equation (1) along the x direction when a phase encoding gradient Gg y is applied yields… the phase-modulated projection of the sensitivity-weighted image onto the x-axis… With reference to FIG. 4, the Fourier Transform of the signals received in all three coils respectively is equal to the projection of the phase encoded image weighted by the three profiles W1(i,j), W2(i,j), W3(i,j) respectively onto the x axis… Assume the image to be of size 3×3 as shown in FIG. 4, where each pixel has an intensity value denoted by r(i,j), where i is the column number and j is the row number. The goal is to determine and reconstruct the values of r(i,j) for all the pixels of the image… in order to reconstruct the image [p(m,n)], a matrix pseudoinverse of [Amj(k,y)] is computed for every position mj along the x-axis. This yields a column by column reconstruction; Col. 10-11: Gx and Gy represent the gradients applied in the x and y directions respectively. Then taking the fourier transform with respect to x provides… the phase modulated projection of the image r (x,y) weighted by the sensitivity profile Wk(i,j) onto the x-axis… In order to appropriately account for all of the frequencies included in the image, the choice of the phase modulations used in the inversion matrix should be determined by the frequency content of the sensitivity profile; at least one of: wherein the MRI information includes a signal collected over a time t and channels q, wherein the inverted column information includes line-intensity profiles, or further comprising:
(a) inverting a further column in the plurality of coil sensitivity weighted projections to generate further inverted column information;
(b) reconstructing at least one further portion of the at least one image based on the further inverted column information; and
(c) repeating procedures (a) and (b) until the at least one image is reconstructed in its entirety (e.g. apparatus and methods for magnetic resonance imaging (MRI) includes reconstructing an image of an object or subject of interest to be examined, such as a patient (i.e. at least one patient), for example, by using an MRI system which includes a plurality of gradient coils that produce spatially encoded gradients imposed upon a static magnetic field within a region (i.e. portion, area, section, etc.) in which the object or subject to be examined is placed (i.e. at least one portion of the at least one image), for example, by acquiring (i.e. receiving, obtaining, etc.) a plurality of magnetic resonance signals of the object or subject of interest and using sensitivity profiles of an array of RF receiver coils at least partially surrounding the object or subject of interest and computing a two-dimensional sensitivity profile for each receiver coil in a multi-dimensional array (i.e. matrix) that is recorded to the memory device, by using phase modulated projections of received magnetic resonance data onto a frequency encoded x-axis, weighted by 2D sensitivity profiles of the coils in the array, for example, and using a matrix pseudoinverse computed for every (i.e. further, another, etc.) position along the x-axis (i.e. (a) inverting a further column in the plurality of coil sensitivity weighted projections to generate further inverted column information), which yields a column by column reconstruction (i.e. (b) reconstructing at least one further portion of the at least one image based on the further inverted column information), including an image of the object or subject of interest reconstructed by the programmed central processor, by combining the inverse of the matrix of the sensitivity profiles of each receiver coil and the matrix of the recorded MR data signals together to provide an image that is displayed or printed (i.e. (c) repeating procedures (a) and (b) until the at least one image is reconstructed in its entirety), as indicated above), for example).
Claims 1-12 are rejected under 35 U.S.C. 103 as being unpatentable over Kyriakos, in view Schlemper et al. (US PG Publication No. 2020/0294287 A1), hereafter referred to as Schlemper, in further view of Levin, and in further view of James.
Regarding claim 1, Kyriakos discloses computer-executable instructions for reconstructing at least one portion of at least one image of at least one patient (Col. 1: apparatus and methods for magnetic resonance imaging, also known as magnetic resonance imaging (MRI)… wherein magnetic resonance data is acquired in parallel using an array of receiver coils at least partially Surrounding the object of interest; Col. 6: An object or subject (not shown) may be placed in the imaging volume within the cylinder 10 for examination using NMR phenomena. The subject is placed on the z-axis or the line 8, and is located within the coil 12. The coil 12 is representative of devices used to generate RF fields in the subject placed in the system for examination), wherein, when a computing arrangement executes the instructions having stored thereon, the computing arrangement is configured to perform procedures (Col. 7: system also includes a central processor 20 which can be a central processing unit (CPU), a memory device 30, and a display device 40… FIG. 2 is a flow diagram illustratively showing the basic sequence of the method of the invention. Thus, it will be seen that the present method contemplates that an apparatus such as that described above will be provided. In this apparatus, the CPU is preprogrammed to carry out numerous operations) comprising:
receiving magnetic resonance imaging (MRI) information for the at least one patient (Col. 1: apparatus and methods for magnetic resonance imaging, also known as magnetic resonance imaging (MRI)… wherein magnetic resonance data is acquired in parallel using an array of receiver coils at least partially Surrounding the object of interest; Col. 2: method generally includes the following steps. First, a magnetic resonance imaging device is provided. This device typically includes a magnet system providing a background magnetic field in an imaging volume, a central processor, a memory device, a RF coil surrounding the imaging volume, and a plurality of radio frequency receiver coils defining a multi-dimensional array thereof disposed about the imaging volume. A two-dimensional sensitivity profile for each receiver coil in the multi-dimensional array is then computed and recorded to the memory device. Thereafter, a plurality of magnetic resonance signals of the object of interest located within the imaging volume is acquired from each receiver coil and recorded to the memory device. An image of the object of interest in a desired plane extending transversely through the imaging volume of the magnetic resonance imaging device then is reconstructed line-by line by the central processor. This reconstruction combines the inverse of the matrix of the sensitivity profiles of each receiver coil and the matrix of the recorded MR data signals together to provide an image that may be displayed or printed; Col. 4: the apparatus preferably constitutes a magnetic resonance imaging device including an imaging volume, preprogrammed central processor, memory device, a body coil surrounding said imaging volume, a plurality of radio frequency receiver coils defining a multi-dimensional array thereof disposed about said imaging volume, and a display device… Each receiver coil in the multi-dimensional array has an associated two-dimensional sensitivity profile… the central processor is adapted to reconstruct, and to display on the display device, line-by-line, a two-dimensional image taken in a selected plane extending transversely through the object of interest; Col. 6: present invention provides a magnetic resonance imaging system. The system includes a plurality of gradient coils that produce spatially encoded gradients imposed upon a background magnetic field Bo within the volume in which the object to be examined is placed… object or subject (not shown) may be placed in the imaging volume within the cylinder 10 for examination using NMR phenomena. The subject is placed on the z-axis or the line 8, and is located within the coil 12. The coil 12 is representative of devices used to generate RF fields in the subject placed in the system for examination; Col. 7: In MRI systems, it is necessary to provide various gradient coils (not shown) for producing spatially encoding gradients that are imposed upon the static magnetic field within the region in which the subject to be examined is placed also are provided… the system also includes a central processor 20 which can be a central processing unit (CPU), a memory device 30, and a display device 40; receive magnetic resonance imaging (MRI) information for the at least one patient (e.g. apparatus (i.e. a system) and methods for magnetic resonance imaging (MRI) include reconstructing an image of an object or subject of interest to be examined, such as a patient (i.e. reconstructing at least one image of at least one patient), for example, by using sensitivity profiles of an array of RF receiver coils at least partially (i.e. at least one portion) surrounding the object or subject of interest (i.e. reconstructing at least one portion of at least one image of at least one patient), for example, by using a programmed central processor (i.e. a computer hardware arrangement), which performs the following operation steps (i.e. comprising: a computer hardware arrangement configured to), including acquiring (i.e. receiving, obtaining, etc.) a plurality of magnetic resonance signals of the object or subject of interest (i.e. receive magnetic resonance imaging (MRI) information for the at least one patient), as indicated above), for example);
generating a plurality of coil sensitivity weighted projections based on the MRI information;
inverting a column in the plurality of coil sensitivity weighted projections to generate inverted column information; and
reconstructing the at least one portion of the at least one image based on the inverted column information (Col. 2-3: parallel encoding in MRI is achieved by using the sensitivity profiles of an array of RF receiver coils at least partially surrounding the object of interest… Wi(x,y) represents the 2D sensitivity profile of the ith coil of the array, and ρ(x, y) is an image slice in a selected (x, y) plane… equation represents a projection of the phase modulated image ρ(x,y) onto the x-axis. Further, this signal can be represented in discrete form… apparatus and method of the present invention use phase modulated projections of the received magnetic resonance data onto the frequency encoded (x-) axis, weighted by the 2D sensitivity profiles of the coils in the array, in order to reconstruct ρ(x,y) column by column (i.e., orthogonal to the x-axis)… A two-dimensional sensitivity profile for each receiver coil in the multi-dimensional array is then computed and recorded to the memory device. Thereafter, a plurality of magnetic resonance signals of the object of interest located within the imaging volume is acquired from each receiver coil and recorded to the memory device. An image of the object of interest in a desired plane extending transversely through the imaging volume of the magnetic resonance imaging device then is reconstructed line-by line by the central processor. This reconstruction combines the inverse of the matrix of the sensitivity profiles of each receiver coil and the matrix of the recorded MR data signals together to provide an image that may be displayed or printed; Col. 4: an object of interest is located in the imaging volume of the magnetic resonance imaging device. Magnetic resonance data from the plane of interest is acquired from each of the RF receiver coils and stored in memory… determining and storing in memory the point-by-point time difference of signal reception by each of the receiver coils. This information may be combined with the sensitivity information in the inverse of the matrix of the sensitivity profile so as to provide a phase compensated, load-weighted, point-by-point, inverse, sensitivity profile matrix for each said receiver coil… The receiver coils are adapted to acquire simultaneously a plurality of magnetic resonance signals from an object of interest located within the imaging volume, and to transfer the same to the memory device… the central processor is adapted to reconstruct, and to display on the display device, line-by-line, a two-dimensional image taken in a selected plane extending transversely through the object of interest…This is accomplished by multiplying the inverse of the matrix of sensitivity profiles of the receiver coils and the matrix of data signals acquired by the receiver coils, and displaying the resultant product on the display device; Col. 8-9: sensitivity profile information from a number of receiver coils is used in order to minimize the number of acquisitions needed to estimate and reconstruct ρ(x, y). Taking the Fourier Transform of equation (1) along the x direction when a phase encoding gradient Gg y is applied yields… the phase-modulated projection of the sensitivity-weighted image onto the x-axis… With reference to FIG. 4, the Fourier Transform of the signals received in all three coils respectively is equal to the projection of the phase encoded image weighted by the three profiles W1(i,j), W2(i,j), W3(i,j) respectively onto the x axis… Assume the image to be of size 3×3 as shown in FIG. 4, where each pixel has an intensity value denoted by r(i,j), where i is the column number and j is the row number… in order to reconstruct the image [p(m,n)], a matrix pseudoinverse of [Amj(k,y)] is computed for every position mj along the x-axis. This yields a column by column reconstruction; Col. 10-11: Gx and Gy represent the gradients applied in the x and y directions respectively. Then taking the fourier transform with respect to x provides… the phase modulated projection of the image r (x,y) weighted by the sensitivity profile Wk(i,j) onto the x-axis… In order to appropriately account for all of the frequencies included in the image, the choice of the phase modulations used in the inversion matrix should be determined by the frequency content of the sensitivity profile; generating a plurality of coil sensitivity weighted projections based on the MRI information;
inverting a column in the plurality of coil sensitivity weighted projections to generate inverted column information; and
reconstructing the at least one portion of the at least one image based on the inverted column information (e.g. apparatus and methods for magnetic resonance imaging (MRI) include reconstructing an image of an object or subject of interest to be examined, such as a patient (i.e. the at least one image), for example, by using sensitivity profiles of an array of RF receiver coils at least partially (i.e. at least one portion) surrounding the object or subject of interest (i.e. reconstructing the at least one portion of the at least one image), for example, by computing a two-dimensional sensitivity profile for each receiver coil in a multi-dimensional array (i.e. matrix) that is recorded to the memory device, by using phase modulated projections of received magnetic resonance data onto a frequency encoded x-axis, weighted by 2D sensitivity profiles of the coils in the array (i.e. generating a plurality of coil sensitivity weighted projections based on the MRI information), for example, and using a matrix pseudoinverse computed for every position along the x-axis (i.e. inverting a column in the plurality of coil sensitivity weighted projections to generate inverted column information), which yields a column by column reconstruction (i.e. reconstructing the at least one portion of the at least one image based on column information), including an image of the object or subject of interest reconstructed by the programmed central processor, for example, by combining the inverse of the matrix of the sensitivity profiles of each receiver coil and the matrix of the recorded MR data signals together to provide an image that is displayed or printed, as indicated above), for example).
Kyriakos teachings above disclose apparatus and methods for magnetic resonance imaging including reconstructing an image of an object or subject of interest to be examined, such as a patient, for example, by using a programmed central processor that performs the steps indicated above, but does not expressly disclose a non-transitory computer-accessible medium having stored thereon computer-executable instructions.
However, Schlemper teaches a non-transitory computer-accessible medium having stored thereon computer-executable instructions (Par. [0008]: at least one non-transitory computer readable storage medium storing processor-executable instructions that, when executed by at least one processor, cause the at least one processor to perform a method for generating magnetic resonance (MR) images of a subject from MR data obtained by a magnetic resonance imaging (MRI) system; Par. [0434-449]: one or more programs comprising instructions that, when executed by controller 2206, cause controller 2206 to control system 2200 to operate in accordance with the pulse sequence, and/or any other suitable information. Information stored in pulse sequences repository 2208 may be stored on one or more non-transitory storage media… To perform any of the functionality described herein, the processor 2610 may execute one or more processor-executable instructions stored in one or more non-transitory computer-readable storage media (e.g., the memory 2620), which may serve as non-transitory computer-readable storage media storing processor-executable instructions for execution by the processor 2610… the performance of processes or methods may utilize program instructions executable by a device (e.g., a computer, a processor, or other device) to perform, or control performance of, the processes or methods. In this respect, various inventive concepts may be embodied as a computer readable storage medium (or multiple computer readable storage media) (e.g., a computer memory, one or more floppy discs, compact discs, optical discs, magnetic tapes, flash memories, circuit configurations in Field Programmable Gate Arrays or other semiconductor devices, or other tangible computer storage medium) encoded with one or more programs that, when executed on one or more computers or other processors, perform methods that implement one or more of the various embodiments described above. The computer readable medium or media can be transportable, such that the program or programs stored thereon can be loaded onto one or more different computers or other processors to implement various ones of the aspects described above… computer readable media may be non-transitory media).
Kyriakos and Schlemper are considered to be analogous art because they pertain to MR image processing applications. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to modify the apparatus and methods for magnetic resonance imaging including reconstructing of an image of an object or subject of interest to be examined (as disclosed by Kyriakos) with a non-transitory computer-accessible medium having stored thereon computer-executable instructions (as taught by Schlemper, Abstract, Par. [0008, 434-449]) to cause at least one processor to perform a method for generating magnetic resonance (MR) images of a subject from MR data obtained by a magnetic resonance imaging (MRI) system and to improve the quality of image reconstruction (Schlemper, Abstract, Par. [0003-8, 233, 313, 404, 419, 421, 434, 447, 449]).
The combination of Kyriakos and Schlemper, as a whole, teaches the non-transitory computer-accessible medium, as indicated above, but fails to teach the following as further recited in claim 1.
However, Levin teaches wherein the MRI information comprises a single echo acquired without requiring a phase encoding ( Par. [0035-41]: method for magnetic resonance imaging is proposed, which is based on transverse magnetization excitation in an additional (reference) slice location and a motion compensation based on a determined reference echo signal phase shift… the at least one reference echo signal is formed without phase encoding; Par. [0053-68]: acquiring the at least one first echo signal with phase encoding and the at least one reference echo signal without phase encoding… a processor means in order to compensate a motion-related phase shift of the at least one echo signal based on the at least one reference phase shift… receiver means is additionally adjusted to acquire the at least one reference echo signal without phase encoding… present invention may use a phase shift of the non-phase encoded reference echo-signal accumulated due to applying the diffusion-weighting gradients in order to characterize bulk motion and tissue deformation and to compensate their effect for adjusting the magnetic resonance image; wherein the MRI information comprises a single echo acquired without requiring a phase encoding (e.g. method for magnetic resonance imaging (MRI) is proposed, in which at least one reference echo signal (i.e. a single echo) is formed (i.e. acquired, obtained, constructed, etc.) without phase encoding (i.e. wherein the MRI information comprises a single echo acquired without requiring a phase encoding), as indicated above), for example).
Kyriakos, Schlemper, and Levin are considered to be analogous art because they pertain to MR image processing applications. Therefore, the combined teachings of Kyriakos, Schlemper, and Levin, as a whole, would have rendered obvious the invention recited in claim 1 with a reasonable expectation of success in order to modify the non-transitory computer-accessible medium for magnetic resonance imaging including reconstructing of an image of an object or subject of interest to be examined (as disclosed by Kyriakos and Schlemper) with wherein the MRI information comprises a single echo acquired without requiring a phase encoding (as taught by Levin, Abstract, Par. [0035-41, 53-68]) to characterize bulk motion and tissue deformation and to compensate their effect for adjusting the magnetic resonance image (Levin, Abstract, Par. [0035-41, 53-68]).
The combination of Kyriakos, Schlemper, and Levin, as a whole, teaches the non-transitory computer-accessible medium, as indicated above, but fails to teach the following as further recited in claim 1.
However, James teaches without requiring a phase encoding gradient for spatial encoding (Par. [0020-21]: Non-invasive techniques that do not rely on use of ionizing radiation or radioactive tracers allow the most leeway for early diagnosis and repeat measurement to monitor disease progression and response to therapy. Magnetic Resonance Imaging (MRI), which provides tunable tissue contrast, is just such a non-invasive technique… embodiments disclosed herein provide a method for pathology assessment employing tissue texture using magnetic resonance (MR) which may be used integrally with an MR imaging technique; Par. [0065-94]: MRI Magnetic Resonance Imaging… MRS Magnetic Resonance Spectroscopy… NMR Nuclear Magnetic Resonance… Several benefits result from acquiring data after the gradient is switched off for single-k-value sampling in a reduced volume (the VOI). By proper pulse sequencing, the echo record window can be designed such that recording begins with the highest k-values of interest, as signal level is highest at the echo peak… In certain embodiments disclosed herein, the approach to k-space filling is to acquire only the set of k-values needed for texture evaluation in the targeted pathology, with data acquired after the gradient is switched off; Par. [0116-122]: Within a given acquisition in standard MR practice, there are M samples which are acquired of the echo… If the entire echo is used to measure one k-value, the receive bandwidth can be adjusted so as to pass the most abundant resonant peaks in the underlying NMR spectrum, and attenuate frequencies above them… Taking a straight MRS spectrum (no structural phase encodes), would yield a spectrum consisting primarily of peaks… the full NMR spectrum may be extracted (without any phase encoding gradients… isochromats of interest can be extracted by acquiring N samples of the echo… Since the echo is being played out with no gradient, the strength of the resulting signal at the Isochromat of interest will correspond to the (complex-valued) k-value coefficient of interest; Par. [0285]: methods disclosed herein entail acquisition within either a single VOI or within multiple, interleaved VOIs, within one TR. Data is acquired without use of a spatial encoding gradient to form an image; without requiring a phase encoding gradient for spatial encoding (e.g. method for pathology assessment employing tissue texture using magnetic resonance (MR) used integrally with an MR imaging (i.e. MRI) technique, for example, includes acquiring data after gradient is switched off for single-k-value sampling in a reduced volume (VOI), for example, and by proper pulse sequencing, an echo record window is designed such that recording begins with highest k-values of interest, as signal level is highest at the echo peak (i.e. a single echo acquired), for example, in which a full Nuclear Magnetic Resonance (NMR) spectrum is extracted (i.e. acquired, obtained, etc.) without any phase encoding gradients (i.e. without requiring a phase encoding gradient), for example, including data acquired without use of a spatial encoding gradient to form an image (i.e. without requiring a phase encoding gradient for spatial encoding), as indicated above), for example).
Kyriakos, Schlemper, Levin, and James are considered to be analogous art because they pertain to MR image processing applications. Therefore, the combined teachings of Kyriakos, Schlemper, Levin, and James, as a whole, would have rendered obvious the invention recited in claim 1 with a reasonable expectation of success in order to modify the non-transitory computer-accessible medium for magnetic resonance imaging including reconstructing of an image of an object or subject of interest to be examined (as disclosed by Kyriakos and Schlemper) with without requiring a phase encoding gradient for spatial encoding (as taught by James, Abstract, Par. [0020-21, 65-94, 116-122, 285]) to circumvent the problem of signal degradation due to patient motion, to provide a method for pathology assessment employing tissue texture using magnetic resonance (MR) which may be used integrally with an MR imaging technique, by acquiring MRI data without use of a spatial encoding gradient to form an image which significantly shortens acquisition time and, combined with a narrowly targeted acquisition in k-space, enables acquisition of requisite data fast enough to provide immunity to subject motion (James, Abstract, Par. [0003-6, 11, 20-21, 65-94, 116-122, 285]).
Regarding claim 2, claim 1 is incorporated and the combination of Kyriakos, Schlemper, Levin, and James, as a whole, teaches the non-transitory computer-accessible medium for magnetic resonance imaging (Kyriakos, Col. 1 and Schlemper, Par. [0008]), wherein the computing arrangement is further configured to deblur the at least one portion of the at least one image (Schlemper, Par. [0007-9]: a system comprising at least one processor configured to perform: obtaining input MR spatial frequency data obtained by imaging the subject using the MRI system; generating an MR image of the subject from the input MR spatial frequency data using a neural network model comprising: a pre-reconstruction neural network configured to process the input MR spatial frequency data; a reconstruction neural network configured to generate at least one initial image of the subject from output of the pre-reconstruction neural network; and a post-reconstruction neural network configured to generate the MR image of the subject from the at least one initial image of the subject… at least one non-transitory computer readable storage medium storing processor-executable instructions that, when executed by at least one processor, cause the at least one processor to perform a method for generating magnetic resonance (MR) images of a subject from MR data obtained by a magnetic resonance imaging (MRI) system. The method comprises: obtaining input MR spatial frequency data obtained by imaging the subject using the MRI system; generating an MR image of the subject from the input MR spatial frequency data using a neural network model comprising: a pre-reconstruction neural network configured to process the input MR spatial frequency data; a reconstruction neural network configured to generate at least one initial image of the subject from output of the pre-reconstruction neural network; and a post-reconstruction neural network configured to generate the MR image of the subject from the at least one initial image of the subject… a method for generating magnetic resonance (MR) images of a subject from MR data obtained by a magnetic resonance imaging (MRI) system. The method comprising: obtaining first input MR data obtained by imaging the subject using the MRI system; obtaining second input MR data obtained by imaging the subject using the MRI system; generating a first set of one or more MR images from the first input MR data; generating a second set of one or more MR images from the second input MR data; aligning the first set of MR images and the second set of MR images using a neural network model to obtain aligned first and second sets of MR images, the neural network model comprising a first neural network and a second neural network, the aligning comprising: estimating, using the first neural network, a first transformation between the first set of MR images and the second set of MR images; generating a first updated set of MR images from the second set of MR images using the first transformation; estimating, using the second neural network, a second transformation between the first set of MR images and the first updated set of MR images; and aligning the first set of MR images and the second set of MR images at least in part by using the first transformation and the second transformation; combining the aligned first and second sets of MR images to obtain a combined set of one or more MR images; and outputting the combined set of one or more MR images; Par. [0078-80]: the reconstruction neural network is configured to perform data consistency processing using a non-uniform Fourier transformation for transforming image data to spatial frequency data… the MRI system comprises a plurality of RF coils, the at least one initial image of the subject comprises a plurality of images, each of the plurality of images generated from a portion of the input MR spatial frequency data collected by a respective RF coil in a plurality of RF coils, and the post-reconstruction neural network comprises a first neural network (e.g., neural network 232) configured to estimate a plurality of RF coil profiles corresponding to the plurality of RF coils… the method further comprises: generating the MR image of the subject using the plurality of MR images and the plurality of RF coil profiles… the at least one initial image of the subject comprises a first set of one or more MR images and a second set of one or more MR images, and the post-reconstruction neural network comprises a second neural network (e.g., neural network 234) for aligning the first set of MR images and the second set of MR images; Par. [0117-123]: Neural network 232 may be used in embodiments in which the MRI system collects data using multiple RF coils… the neural network 232 may be used to combine the images (from among initial images 232) generated from data collected by different RF coils, but corresponding to the same slices…neural network 232 may be used to either estimate such a combined image directly or to estimate sensitivity profiles for the different RF coils, which in turn may be used to combine the images… neural network 234 may be configured to align two sets of one or more MR images to each other. In some instances, each set of MR images may correspond to a set of images for a given volume (e.g., a number of 2D slices that may be stacked to constitute a volume). Such an alignment allows for the sets of MR images to be averaged to increase the SNR. Performing the averaging without first performing alignment would introduce blurring due to, for example, movement of the patient during acquisition of the data being averaged… the neural network 236 may be applied after neural network 234 is used to align corresponding sets of images so that blurring is not introduced through the combination performed by neural network 236; wherein the computing arrangement is further configured to deblur the at least one portion of the at least one image (e.g. system and method for generating magnetic resonance (MR) images of a subject from MR data obtained by a magnetic resonance imaging (MRI) system, in which a plurality of images is generated from a portion of input MR spatial frequency data collected by a respective RF coil in a plurality of RF coils of the MRI system (i.e. at least one portion of the at least one image), for example, including at least one processor configured to (i.e. the computing arrangement is further configured to) implement a neural network used to align corresponding sets of images, including a plurality of images generated from a portion of the input MR spatial frequency data collected, for example, so that blurring is not introduced (i.e. deblur, prevent blurring, etc.) through the combination performed by the neural network used to align the corresponding sets of images (i.e. wherein the computing arrangement is further configured to deblur the at least one portion of the at least one image), as indicated above), for example).
The same motivation to combine above-mentioned teachings applies, as previously indicated in claim 1.
Regarding claim 3, claim 2 is incorporated and the combination of Kyriakos, Schlemper, Levin, and James, as a whole, teaches the non-transitory computer-accessible medium for magnetic resonance imaging (Kyriakos, Col. 1 and Schlemper, Par. [0008]), wherein the computing arrangement is configured to deblur the at least one portion of the at least one image using at least one deep learning procedure (Schlemper, Par. [0007-9]: a system comprising at least one processor configured to perform: obtaining input MR spatial frequency data obtained by imaging the subject using the MRI system; generating an MR image of the subject from the input MR spatial frequency data using a neural network model comprising: a pre-reconstruction neural network configured to process the input MR spatial frequency data; a reconstruction neural network configured to generate at least one initial image of the subject from output of the pre-reconstruction neural network; and a post-reconstruction neural network configured to generate the MR image of the subject from the at least one initial image of the subject… at least one non-transitory computer readable storage medium storing processor-executable instructions that, when executed by at least one processor, cause the at least one processor to perform a method for generating magnetic resonance (MR) images of a subject from MR data obtained by a magnetic resonance imaging (MRI) system. The method comprises: obtaining input MR spatial frequency data obtained by imaging the subject using the MRI system; generating an MR image of the subject from the input MR spatial frequency data using a neural network model comprising: a pre-reconstruction neural network configured to process the input MR spatial frequency data; a reconstruction neural network configured to generate at least one initial image of the subject from output of the pre-reconstruction neural network; and a post-reconstruction neural network configured to generate the MR image of the subject from the at least one initial image of the subject… a method for generating magnetic resonance (MR) images of a subject from MR data obtained by a magnetic resonance imaging (MRI) system. The method comprising: obtaining first input MR data obtained by imaging the subject using the MRI system; obtaining second input MR data obtained by imaging the subject using the MRI system; generating a first set of one or more MR images from the first input MR data; generating a second set of one or more MR images from the second input MR data; aligning the first set of MR images and the second set of MR images using a neural network model to obtain aligned first and second sets of MR images, the neural network model comprising a first neural network and a second neural network, the aligning comprising: estimating, using the first neural network, a first transformation between the first set of MR images and the second set of MR images; generating a first updated set of MR images from the second set of MR images using the first transformation; estimating, using the second neural network, a second transformation between the first set of MR images and the first updated set of MR images; and aligning the first set of MR images and the second set of MR images at least in part by using the first transformation and the second transformation; combining the aligned first and second sets of MR images to obtain a combined set of one or more MR images; and outputting the combined set of one or more MR images; Par. [0078-80]: the reconstruction neural network is configured to perform data consistency processing using a non-uniform Fourier transformation for transforming image data to spatial frequency data… the MRI system comprises a plurality of RF coils, the at least one initial image of the subject comprises a plurality of images, each of the plurality of images generated from a portion of the input MR spatial frequency data collected by a respective RF coil in a plurality of RF coils, and the post-reconstruction neural network comprises a first neural network (e.g., neural network 232) configured to estimate a plurality of RF coil profiles corresponding to the plurality of RF coils… the method further comprises: generating the MR image of the subject using the plurality of MR images and the plurality of RF coil profiles… the at least one initial image of the subject comprises a first set of one or more MR images and a second set of one or more MR images, and the post-reconstruction neural network comprises a second neural network (e.g., neural network 234) for aligning the first set of MR images and the second set of MR images; Par. [0117-123]: Neural network 232 may be used in embodiments in which the MRI system collects data using multiple RF coils… the neural network 232 may be used to combine the images (from among initial images 232) generated from data collected by different RF coils, but corresponding to the same slices…neural network 232 may be used to either estimate such a combined image directly or to estimate sensitivity profiles for the different RF coils, which in turn may be used to combine the images… neural network 234 may be configured to align two sets of one or more MR images to each other. In some instances, each set of MR images may correspond to a set of images for a given volume (e.g., a number of 2D slices that may be stacked to constitute a volume). Such an alignment allows for the sets of MR images to be averaged to increase the SNR. Performing the averaging without first performing alignment would introduce blurring due to, for example, movement of the patient during acquisition of the data being averaged… the neural network 236 may be applied after neural network 234 is used to align corresponding sets of images so that blurring is not introduced through the combination performed by neural network 236; Par. [0219-220]: inventors have developed deep learning techniques for aligning sets of images obtained by multiple acquisitions of the same slice and/or volume… the deep learning techniques involve using a cascade of two or more neural networks configured to estimate a transformation (e.g., a non-rigid, an affine, a rigid transformation) between two sets of MR images (each set having one or multiple MR images), and aligning the two sets of images using the estimated transformation. In turn, the two sets of images may be averaged to obtain a combined set of images having a higher SNR than the sets of images themselves… the estimated transformation may indicate one or more rotations and/or translations to align the two sets of images… the deep learning techniques described herein may be used as part of neural network 234 part of post-reconstruction neural network 214, as described herein including in connection with FIG. 2C; wherein the computing arrangement is configured to deblur the at least one portion of the at least one image using at least one deep learning procedure (e.g. system and method for generating magnetic resonance (MR) images of a subject from MR data obtained by a magnetic resonance imaging (MRI) system, in which a plurality of images is generated from a portion of input MR spatial frequency data collected by a respective RF coil in a plurality of RF coils of the MRI system (i.e. at least one portion of the at least one image), for example, including at least one processor configured to (i.e. the computing arrangement is further configured to) implement a neural network used to align corresponding sets of images, including the plurality of images generated from a portion of the input MR spatial frequency data collected, so that blurring is not introduced (i.e. deblur, prevent blurring, etc.) through the combination performed by the neural network used to align the corresponding sets of images, for example, including deep learning techniques for aligning sets of images obtained by multiple acquisitions of the same slice and/or volume (i.e. wherein the computing arrangement is configured to deblur the at least one portion of the at least one image using at least one deep learning procedure), as indicated above), for example).
The same motivation to combine above-mentioned teachings applies, as previously indicated in claim 1.
Regarding claim 4, claim 3 is incorporated and the combination of Kyriakos, Schlemper, Levin, and James, as a whole, teaches the non-transitory computer-accessible medium for magnetic resonance imaging (Kyriakos, Col. 1 and Schlemper, Par. [0008]), wherein the computing arrangement is further configured to:
receive a reference scan of at least one part of the at least one patient; and
train the at least one deep learning procedure based on the reference scan (Schlemper, Par. [0186-216]: neural network models described herein may be trained using any suitable neural network training algorithm(s), as aspects of the technology described herein are not limited in this respect… the neural network models described herein may be trained by using one or more iterative optimization techniques to estimate neural network parameters from training data… training data for training a neural network may be generated synthetically from available MR images… magnitude MR images (phase information is typically discarded) may be used to generate corresponding spatial frequency data and the resulting (spatial frequency data, MR image) pairs may be used to train a neural network model, including any of the neural network models described herein… Using characteristics of the MRI system that will collect patient data to generate training data allows for the neural network to learn these characteristics and use them to improve its performance on tasks in the reconstruction pipeline. Moreover, this approach allows the trained neural network models to reconstruct MR images of comparably high quality based on sensor data acquired using MRI hardware and software… a process 500 for generating training data from MR images for training the neural network models described herein… The process 500 starts with a magnitude MR volume 502 using various specified characteristics of an MRI system generates spatial frequency data 550, which includes spatial frequency data collected multiple times… by each of multiple RF colis of the MRI system… Process 500 may be performed by any suitable computing device(s)… process 500 begins by accessing a reference magnitude MR volume 502. The MR volume 502 may comprise one or multiple images. Each of the image(s) may represent an anatomical slice of a subject being imaged. The MR volume 502 may include one or more magnitude images obtained by a clinical MRI system… neural network models developed by the inventors and described herein may be trained using training data generated from existing high-field image data… a training dataset of (sensor input data, image) pairs may be generated by, for each pair, starting with a high-field source image xh … training the neural network with data pairs derived from high-field data (as above), but also augmenting the loss function with losses computed with respect to available low-field images. The key insight is that, even if a neural network were trained using high-field data, the resulting network should reconstruct the same image from both: (1) a first set of low-field k-space data; and (2) a second set of low-field data obtained by applying a geometric transformation to the first set of low-field k-space data, where the image reconstruction should be invariant under the transformation… another way to generate a training dataset is to use source images of higher quality xo, such as those obtained from low-field scanners, but using more data samples. The sensor data can be obtained directly by collecting the scanner measurements yo; Par. [0303]: additional aspects of training neural networks configured to perform motion estimation and/or correction are described… It may, in some instances, be difficult to acquire large scale real motion-corrupted data for training of any of the neural network models described herein. Accordingly… it may be desirable to generate synthetic training data including reference MR images and synthetic motion-corrupted MR images based on existing datasets 1002 of MR images; Par. [0389]: training data for training a neural network for estimating coil profiles may be generated synthetically from a dataset of existing MR scans. For example… an MR image x may be loaded from a dataset and random phase may be added to this image to obtain a complex-valued image (since only magnitudes are typically available in existing datasets). Complex-valued coil profiles Si for Ncoil coils may be synthesized next. For example, the sensitivity values for particular pixels/voxels may be sampled according to a Gaussian distribution and random phase may be added; wherein the computing arrangement is further configured to:
receive a reference scan of at least one part of the at least one patient; and
train the at least one deep learning procedure based on the reference scan (e.g. system and method for generating magnetic resonance (MR) images of a subject (i.e. at least one patient) from MR data obtained by a magnetic resonance imaging (MRI) system, in which a plurality of images is generated from a portion of input MR spatial frequency data collected by a respective RF coil in a plurality of RF coils of the MRI system (i.e. at least one part of the at least one patient), for example, including at least one processor configured to (i.e. the computing arrangement is further configured to) implement a neural network using deep learning techniques (i.e. at least one deep learning procedure), including obtaining (i.e. receiving) training data for training a neural network (i.e. train the at least one deep learning procedure) for estimating coil profiles, which is generated synthetically from a dataset of existing (i.e. reference, source, etc.) MR scans (i.e. receive a reference scan of at least one part of the at least one patient), for example, by using characteristics of the MRI system that will collect subject or patient data to generate training data (i.e. receive a reference scan of at least one part of the at least one patient; and train the at least one deep learning procedure based on the reference scan), as indicated above), for example).
The same motivation to combine above-mentioned teachings applies, as previously indicated in claim 1.
Regarding claim 5, claim 4 is incorporated and the combination of Kyriakos, Schlemper, Levin, and James, as a whole, teaches the non-transitory computer-accessible medium for magnetic resonance imaging (Kyriakos, Col. 1 and Schlemper, Par. [0008]), wherein the computing arrangement is further configured to:
generate a plurality of training images by varying at least one of (i) an amplitude of the reference scan, or (ii) a noise level of the reference scan; and
train the at least one deep learning procedure based on the plurality of training images (Schlemper, Par. [0186-216]: neural network models described herein may be trained using any suitable neural network training algorithm(s), as aspects of the technology described herein are not limited in this respect… the neural network models described herein may be trained by using one or more iterative optimization techniques to estimate neural network parameters from training data… training data for training a neural network may be generated synthetically from available MR images… magnitude MR images (phase information is typically discarded) may be used to generate corresponding spatial frequency data and the resulting (spatial frequency data, MR image) pairs may be used to train a neural network model, including any of the neural network models described herein… Using characteristics of the MRI system that will collect patient data to generate training data allows for the neural network to learn these characteristics and use them to improve its performance on tasks in the reconstruction pipeline. Moreover, this approach allows the trained neural network models to reconstruct MR images of comparably high quality based on sensor data acquired using MRI hardware and software… a process 500 for generating training data from MR images for training the neural network models described herein… The process 500 starts with a magnitude MR volume 502 using various specified characteristics of an MRI system generates spatial frequency data 550, which includes spatial frequency data collected multiple times… by each of multiple RF colis of the MRI system… Process 500 may be performed by any suitable computing device(s)… process 500 begins by accessing a reference magnitude MR volume 502. The MR volume 502 may comprise one or multiple images. Each of the image(s) may represent an anatomical slice of a subject being imaged. The MR volume 502 may include one or more magnitude images obtained by a clinical MRI system… an RF coil sensitivity profile is generated for each of the Ncoil RF coils to obtain multiple RF coil sensitivity profiles… Each generated RF coil sensitivity profile Si is complex-valued, with the magnitudes generated at act 526 using one or more RF coil models and with the phases generated (e.g., randomly) at 528. The resulting RF sensitivity profiles are applied to the MR volume (e.g., to the result of performing, at 524, pulse sequence specific augmentation on target MR volume 520) to obtain multiple MR volumes, each of the multiple MR volumes obtained by applying a respective RF coil sensitivity profile to the MR volume resulting at the output of 524… at 538 and 540, correlated Gaussian noise is generated and added, at 542, to the multiple MR volumes produced at 536… the Gaussian noise may be generated by: (1) determining, at 538, a noise level σi for each of the coils; and (2) generating, at 540, Gaussian noise having the covariance of LDLT, where D is a diagonal matrix… and L is the coil correlation matrix determined at 534… at 544, a k-space sampling trajectory is selected… Next, at 546, noise δk(t) is added to sampling trajectory k(t). The noise may be added to simulate for various MRI system imperfections and/or any other reason. Next, at 548, a non-uniform Fourier transform is applied to the noise-corrupted coil-weighted MR volumes produced at 542… The resulting spatial frequency data are then output, at 550. These data may be used for training any of the neural network models described herein…neural network models developed by the inventors and described herein may be trained using training data generated from existing high-field image data… a training dataset of (sensor input data, image) pairs may be generated by, for each pair, starting with a high-field source image xh … training the neural network with data pairs derived from high-field data (as above), but also augmenting the loss function with losses computed with respect to available low-field images. The key insight is that, even if a neural network were trained using high-field data, the resulting network should reconstruct the same image from both: (1) a first set of low-field k-space data; and (2) a second set of low-field data obtained by applying a geometric transformation to the first set of low-field k-space data, where the image reconstruction should be invariant under the transformation… another way to generate a training dataset is to use source images of higher quality xo, such as those obtained from low-field scanners, but using more data samples. The sensor data can be obtained directly by collecting the scanner measurements yo; Par. [0303-308]: additional aspects of training neural networks configured to perform motion estimation and/or correction are described… It may, in some instances, be difficult to acquire large scale real motion-corrupted data for training of any of the neural network models described herein. Accordingly… it may be desirable to generate synthetic training data including reference MR images and synthetic motion-corrupted MR images based on existing datasets 1002 of MR images… To generate such synthetic training datasets, a volume may be selected and loaded in act 1004 from dataset 1002… only a magnitude portion of the volume may be loaded. After loading the selected volume in act 1004, a random affine transformation matrix T may be sampled in act 1006… The transformed volume may be stored as a reference volume… To better train the neural network model, it may be desirable to include synthetic noise in the synthetic training data (e.g., to simulate non-ideal MR imaging conditions). In act 1014, Gaussian noise may be sampled in act 1014. The Gaussian noise may be selected to match the volume size of the loaded volume… noise may be added to the reference volume and the moving volume by undersampling a percentage of the MR data in k-space. In act 1016, the Gaussian noise may be added to the reference volume and the moving volume to form the synthetic training data pair for use by the neural network model; Par. [0389]: training data for training a neural network for estimating coil profiles may be generated synthetically from a dataset of existing MR scans. For example… an MR image x may be loaded from a dataset and random phase may be added to this image to obtain a complex-valued image (since only magnitudes are typically available in existing datasets). Complex-valued coil profiles Si for Ncoil coils may be synthesized next. For example, the sensitivity values for particular pixels/voxels may be sampled according to a Gaussian distribution and random phase may be added; wherein the computing arrangement is further configured to:
generate a plurality of training images by varying at least one of (i) an amplitude of the reference scan, or (ii) a noise level of the reference scan; and
train the at least one deep learning procedure based on the plurality of training images (e.g. system and method for generating magnetic resonance (MR) images of a subject (i.e. at least one patient) from MR data obtained by a magnetic resonance imaging (MRI) system, in which a plurality of images is generated from a portion of input MR spatial frequency data collected by a respective RF coil in a plurality of RF coils of the MRI system, for example, including at least one processor configured to (i.e. the computing arrangement is further configured to) implement a neural network using deep learning techniques (i.e. at least one deep learning procedure), including obtaining (i.e. receiving) training data for training a neural network for estimating coil profiles, which is generated synthetically from a dataset of existing (i.e. reference, source, etc.) MR scans (i.e. MR images), for example, by using characteristics of the MRI system that will collect subject or patient data to generate training data (i.e. generate a plurality of training images), including (i.e. at least one of) adding noise to a reference volume (i.e. varying at least one of a noise level of the reference scan) to simulate for various MRI system imperfections, and/or any other reason, for example, in order to form the synthetic training data for use by the neural network during training (i.e. and train the at least one deep learning procedure based on the plurality of training images), as indicated above), for example).
The same motivation to combine above-mentioned teachings applies, as previously indicated in claim 1.
Regarding claim 6, claim 1 is incorporated and the combination of Kyriakos, Schlemper, Levin, and James, as a whole, teaches the non-transitory computer-accessible medium for magnetic resonance imaging (Kyriakos, Col. 1 and Schlemper, Par. [0008]), wherein the computing arrangement is further configured to:
(a) invert a further column in the plurality of coil sensitivity weighted projections to generate further inverted column information;
(b) reconstruct at least one further portion of the at least one image based on the further inverted column information; and
(c) repeat procedures (a) and (b) until the at least one image is reconstructed in its entirety (Kyriakos, Col. 2-3: parallel encoding in MRI is achieved by using the sensitivity profiles of an array of RF receiver coils at least partially surrounding the object of interest… Wi(x,y) represents the 2D sensitivity profile of the ith coil of the array, and ρ(x, y) is an image slice in a selected (x, y) plane… equation represents a projection of the phase modulated image ρ(x,y) onto the x-axis. Further, this signal can be represented in discrete form… apparatus and method of the present invention use phase modulated projections of the received magnetic resonance data onto the frequency encoded (x-) axis, weighted by the 2D sensitivity profiles of the coils in the array, in order to reconstruct ρ(x,y) column by column (i.e., orthogonal to the x-axis)… A two-dimensional sensitivity profile for each receiver coil in the multi-dimensional array is then computed and recorded to the memory device. Thereafter, a plurality of magnetic resonance signals of the object of interest located within the imaging volume is acquired from each receiver coil and recorded to the memory device. An image of the object of interest in a desired plane extending transversely through the imaging volume of the magnetic resonance imaging device then is reconstructed line-by line by the central processor. This reconstruction combines the inverse of the matrix of the sensitivity profiles of each receiver coil and the matrix of the recorded MR data signals together to provide an image that may be displayed or printed; Col. 4: an object of interest is located in the imaging volume of the magnetic resonance imaging device. Magnetic resonance data from the plane of interest is acquired from each of the RF receiver coils and stored in memory… determining and storing in memory the point-by-point time difference of signal reception by each of the receiver coils. This information may be combined with the sensitivity information in the inverse of the matrix of the sensitivity profile so as to provide a phase compensated, load-weighted, point-by-point, inverse, sensitivity profile matrix for each said receiver coil… The receiver coils are adapted to acquire simultaneously a plurality of magnetic resonance signals from an object of interest located within the imaging volume, and to transfer the same to the memory device… the central processor is adapted to reconstruct, and to display on the display device, line-by-line, a two-dimensional image taken in a selected plane extending transversely through the object of interest…This is accomplished by multiplying the inverse of the matrix of sensitivity profiles of the receiver coils and the matrix of data signals acquired by the receiver coils, and displaying the resultant product on the display device; Col. 8-9: sensitivity profile information from a number of receiver coils is used in order to minimize the number of acquisitions needed to estimate and reconstruct ρ(x, y). Taking the Fourier Transform of equation (1) along the x direction when a phase encoding gradient Gg y is applied yields… the phase-modulated projection of the sensitivity-weighted image onto the x-axis… With reference to FIG. 4, the Fourier Transform of the signals received in all three coils respectively is equal to the projection of the phase encoded image weighted by the three profiles W1(i,j), W2(i,j), W3(i,j) respectively onto the x axis… Assume the image to be of size 3×3 as shown in FIG. 4, where each pixel has an intensity value denoted by r(i,j), where i is the column number and j is the row number. The goal is to determine and reconstruct the values of r(i,j) for all the pixels of the image… in order to reconstruct the image [p(m,n)], a matrix pseudoinverse of [Amj(k,y)] is computed for every position mj along the x-axis. This yields a column by column reconstruction; Col. 10-11: Gx and Gy represent the gradients applied in the x and y directions respectively. Then taking the fourier transform with respect to x provides… the phase modulated projection of the image r (x,y) weighted by the sensitivity profile Wk(i,j) onto the x-axis… In order to appropriately account for all of the frequencies included in the image, the choice of the phase modulations used in the inversion matrix should be determined by the frequency content of the sensitivity profile; wherein the computing arrangement is further configured to:
(a) invert a further column in the plurality of coil sensitivity weighted projections to generate further inverted column information;
(b) reconstruct at least one further portion of the at least one image based on the further inverted column information; and
(c) repeat procedures (a) and (b) until the at least one image is reconstructed in its entirety (e.g. apparatus and methods for magnetic resonance imaging (MRI) includes reconstructing an image of an object or subject of interest to be examined, such as a patient (i.e. at least one patient), for example, by using an MRI system which includes a plurality of gradient coils that produce spatially encoded gradients imposed upon a static magnetic field within a region (i.e. portion, area, section, etc.) in which the object or subject to be examined is placed (i.e. at least one portion of the at least one image), for example, by acquiring (i.e. receiving, obtaining, etc.) a plurality of magnetic resonance signals of the object or subject of interest and using sensitivity profiles of an array of RF receiver coils at least partially surrounding the object or subject of interest and computing a two-dimensional sensitivity profile for each receiver coil in a multi-dimensional array (i.e. matrix) that is recorded to the memory device, by using phase modulated projections of received magnetic resonance data onto a frequency encoded x-axis, weighted by 2D sensitivity profiles of the coils in the array, for example, and using a matrix pseudoinverse computed for every (i.e. further, another, etc.) position along the x-axis (i.e. (a) invert a further column in the plurality of coil sensitivity weighted projections to generate further inverted column information), which yields a column by column reconstruction (i.e. (b) reconstruct at least one further portion of the at least one image based on the further inverted column information), including an image of the object or subject of interest reconstructed by the programmed central processor, by combining the inverse of the matrix of the sensitivity profiles of each receiver coil and the matrix of the recorded MR data signals together to provide an image that is displayed or printed (i.e. (c) repeat procedures (a) and (b) until the at least one image is reconstructed in its entirety), as indicated above), for example).
Regarding claim 7, claim 1 is incorporated and the combination of Kyriakos, Schlemper, Levin, and James, as a whole, teaches the non-transitory computer-accessible medium for magnetic resonance imaging (Kyriakos, Col. 1 and Schlemper, Par. [0008]), wherein the MRI information includes a signal collected over a time t and channels q (Schlemper, Par. [102-202]: the neural network 224 may be trained using particular loss functions described next… An MRI system may have one or multiple RF coils configured to detect MR signals in the imaging region of the MR system… the neural network 224 may be trained to suppress RF interference ic. To this end, training data may be created that includes all of the components of sc separately so that ground truth is available… input to the neural network 224 may be: (1) the signal sc for each coil, so that the neural network suppresses RF interference for each coil separately; (2) the signals sc for all the coils as separate channels, so that the neural network suppresses RF interference for all coils at the same time; or (3) the signals sc for each coil, as separate channels… Neural network 232 may be used in embodiments in which the MRI system collects data using multiple RF coils… the neural network 232 may be used to combine the images (from among initial images 232) generated from data collected by different RF coils, but corresponding to the same slices. As described in more detail below in the “Coil Estimation” Section below, neural network 232 may be used to either estimate such a combined image directly or to estimate sensitivity profiles for the different RF coils, which in turn may be used to combine the images… the process of synthesizing spatial frequency data from MR image data for training a neural network may take into account one or more characteristics of MRI system that will collect patient data that the neural network is being trained to process once the neural network is deployed. Non-limiting, examples of such characteristics include, but are not limited to… sensitivity of RF coils in the MRI system… a process 500 for generating training data from MR images for training the neural network models described herein, in accordance with some embodiments of the technology described herein. The process 500 starts with a magnitude MR volume 502 using various specified characteristics of an MRI system generates spatial frequency data 550, which includes spatial frequency data collected multiple times (Navg times in this example) by each of multiple RF cols of the MRI system… process 500 may be repeated multiple times by starting from the same MR volume 502 to generate different spatial frequency data 550, since multiple portions of the process 500 can be made to vary across different runs since these portions sample certain variations and parameters at random… at acts 526 and 528, an RF coil sensitivity profile is generated for each of the Ncoil RF coils to obtain multiple RF coil sensitivity profiles… Each generated RF coil sensitivity profile Si is complex-valued, with the magnitudes generated at act 526 using one or more RF coil models and with the phases generated (e.g., randomly) at 528. The resulting RF sensitivity profiles are applied to the MR volume (e.g., to the result of performing, at 524, pulse sequence specific augmentation on target MR volume 520) to obtain multiple MR volumes, each of the multiple MR volumes obtained by applying a respective RF coil sensitivity profile to the MR volume resulting at the output of 524; wherein the MRI information includes a signal collected over a time t and channels q (e.g. system and method for generating magnetic resonance (MR) images of a subject from MR data obtained by a magnetic resonance imaging (MRI) system, in which a plurality of images is generated from a portion of input MR spatial frequency data collected by a respective RF coil in a plurality of RF coils of the MRI system, for example, including at least one processor configured to perform functions, including spatial frequency data collected multiple times (i.e. a signal collected over a time t) by each of multiple RF coils of the MRI system, such as signals sc for each coil, for example, which are collected as separate channels (i.e. wherein the MRI information includes a signal collected over a time t and channels q), as indicated above), for example).
The same motivation to combine above-mentioned teachings applies, as previously indicated in claim 1.
Regarding claim 8, claim 7 is incorporated and the combination of Kyriakos, Schlemper, Levin, and James, as a whole, teaches the non-transitory computer-accessible medium for magnetic resonance imaging (Kyriakos, Col. 1 and Schlemper, Par. [0008]), wherein the signal includes a coil sensitivity for each location of each of the channels q (Schlemper, Par. [102-202]: the neural network 224 may be trained using particular loss functions described next… An MRI system may have one or multiple RF coils configured to detect MR signals in the imaging region of the MR system… the neural network 224 may be trained to suppress RF interference ic. To this end, training data may be created that includes all of the components of sc separately so that ground truth is available… input to the neural network 224 may be: (1) the signal sc for each coil, so that the neural network suppresses RF interference for each coil separately; (2) the signals sc for all the coils as separate channels, so that the neural network suppresses RF interference for all coils at the same time; or (3) the signals sc for each coil, as separate channels… Neural network 232 may be used in embodiments in which the MRI system collects data using multiple RF coils… the neural network 232 may be used to combine the images (from among initial images 232) generated from data collected by different RF coils, but corresponding to the same slices. As described in more detail below in the “Coil Estimation” Section below, neural network 232 may be used to either estimate such a combined image directly or to estimate sensitivity profiles for the different RF coils, which in turn may be used to combine the images… the process of synthesizing spatial frequency data from MR image data for training a neural network may take into account one or more characteristics of MRI system that will collect patient data that the neural network is being trained to process once the neural network is deployed. Non-limiting, examples of such characteristics include, but are not limited to… sensitivity of RF coils in the MRI system… a process 500 for generating training data from MR images for training the neural network models described herein, in accordance with some embodiments of the technology described herein. The process 500 starts with a magnitude MR volume 502 using various specified characteristics of an MRI system generates spatial frequency data 550, which includes spatial frequency data collected multiple times (Navg times in this example) by each of multiple RF cols of the MRI system… process 500 may be repeated multiple times by starting from the same MR volume 502 to generate different spatial frequency data 550, since multiple portions of the process 500 can be made to vary across different runs since these portions sample certain variations and parameters at random… at acts 526 and 528, an RF coil sensitivity profile is generated for each of the Ncoil RF coils to obtain multiple RF coil sensitivity profiles… Each generated RF coil sensitivity profile Si is complex-valued, with the magnitudes generated at act 526 using one or more RF coil models and with the phases generated (e.g., randomly) at 528. The resulting RF sensitivity profiles are applied to the MR volume (e.g., to the result of performing, at 524, pulse sequence specific augmentation on target MR volume 520) to obtain multiple MR volumes, each of the multiple MR volumes obtained by applying a respective RF coil sensitivity profile to the MR volume resulting at the output of 524… The RF coil model used at 524 may be of any suitable type… the RF coil model used at 526 may be a physics-based RF coil model, which may be configured to calculate the sensitivity of a particular RF coil given its geometry; wherein the signal includes a coil sensitivity for each location of each of the channels q (e.g. system and method for generating magnetic resonance (MR) images of a subject from MR data obtained by a magnetic resonance imaging (MRI) system, in which a plurality of images is generated from a portion of input MR spatial frequency data collected by a respective RF coil in a plurality of RF coils of the MRI system, for example, including at least one processor configured to perform functions, including spatial frequency data collected multiple times (Navg times in this example) by each of multiple RF coils of the MRI system and signals sc for each coil, as separate channels (i.e. each location of each of the channels q), for example, including a neural network used to either estimate a combined image directly or to estimate sensitivity profiles for different RF coils, which in turn are used to combine the images (i.e. wherein the signal includes a coil sensitivity for each location of each of the channels q), as indicated above), for example).
The same motivation to combine above-mentioned teachings applies, as previously indicated in claim 1.
Regarding claim 9, claim 8 is incorporated and the combination of Kyriakos, Schlemper, Levin, and James, as a whole, teaches the non-transitory computer-accessible medium for magnetic resonance imaging (Kyriakos, Col. 1 and Schlemper, Par. [0008]), wherein the plurality of coil sensitivity weighted projections are generated using a discrete Fourier transform of the signal (Kyriakos, Col. 2-3: parallel encoding in MRI is achieved by using the sensitivity profiles of an array of RF receiver coils at least partially surrounding the object of interest… Wi(x,y) represents the 2D sensitivity profile of the ith coil of the array, and ρ(x, y) is an image slice in a selected (x, y) plane… equation represents a projection of the phase modulated image ρ(x,y) onto the x-axis. Further, this signal can be represented in discrete form… apparatus and method of the present invention use phase modulated projections of the received magnetic resonance data onto the frequency encoded (x-) axis, weighted by the 2D sensitivity profiles of the coils in the array, in order to reconstruct ρ(x,y) column by column (i.e., orthogonal to the x-axis)… A two-dimensional sensitivity profile for each receiver coil in the multi-dimensional array is then computed and recorded to the memory device. Thereafter, a plurality of magnetic resonance signals of the object of interest located within the imaging volume is acquired from each receiver coil and recorded to the memory device. An image of the object of interest in a desired plane extending transversely through the imaging volume of the magnetic resonance imaging device then is reconstructed line-by line by the central processor. This reconstruction combines the inverse of the matrix of the sensitivity profiles of each receiver coil and the matrix of the recorded MR data signals together to provide an image that may be displayed or printed; Col. 4: an object of interest is located in the imaging volume of the magnetic resonance imaging device. Magnetic resonance data from the plane of interest is acquired from each of the RF receiver coils and stored in memory… determining and storing in memory the point-by-point time difference of signal reception by each of the receiver coils. This information may be combined with the sensitivity information in the inverse of the matrix of the sensitivity profile so as to provide a phase compensated, load-weighted, point-by-point, inverse, sensitivity profile matrix for each said receiver coil… The receiver coils are adapted to acquire simultaneously a plurality of magnetic resonance signals from an object of interest located within the imaging volume, and to transfer the same to the memory device… the central processor is adapted to reconstruct, and to display on the display device, line-by-line, a two-dimensional image taken in a selected plane extending transversely through the object of interest…This is accomplished by multiplying the inverse of the matrix of sensitivity profiles of the receiver coils and the matrix of data signals acquired by the receiver coils, and displaying the resultant product on the display device; Col. 8-9: sensitivity profile information from a number of receiver coils is used in order to minimize the number of acquisitions needed to estimate and reconstruct ρ(x, y). Taking the Fourier Transform of equation (1) along the x direction when a phase encoding gradient Gg y is applied yields… the phase-modulated projection of the sensitivity-weighted image onto the x-axis… With reference to FIG. 4, the Fourier Transform of the signals received in all three coils respectively is equal to the projection of the phase encoded image weighted by the three profiles W1(i,j), W2(i,j), W3(i,j) respectively onto the x axis… Assume the image to be of size 3×3 as shown in FIG. 4, where each pixel has an intensity value denoted by r(i,j), where i is the column number and j is the row number. The goal is to determine and reconstruct the values of r(i,j) for all the pixels of the image… in order to reconstruct the image [p(m,n)], a matrix pseudoinverse of [Amj(k,y)] is computed for every position mj along the x-axis. This yields a column by column reconstruction; Col. 10-11: Gx and Gy represent the gradients applied in the x and y directions respectively. Then taking the fourier transform with respect to x provides… the phase modulated projection of the image r (x,y) weighted by the sensitivity profile Wk(i,j) onto the x-axis… In order to appropriately account for all of the frequencies included in the image, the choice of the phase modulations used in the inversion matrix should be determined by the frequency content of the sensitivity profile; wherein the plurality of coil sensitivity weighted projections are generated using a discrete Fourier transform of the signal (e.g. apparatus and methods for magnetic resonance imaging (MRI) includes reconstructing an image of an object or subject of interest to be examined, such as a patient, for example, by using an MRI system which includes a plurality of gradient coils that produce spatially encoded gradients imposed upon a static magnetic field within a region in which the object or subject to be examined is placed, for example, acquiring a plurality of magnetic resonance signals of the object or subject of interest and using sensitivity profiles of an array of RF receiver coils at least partially surrounding the object or subject of interest, and computing a two-dimensional sensitivity profile for each receiver coil in a multi-dimensional array that is recorded to the memory device, by using phase modulated projections of received magnetic resonance data onto a frequency encoded x-axis, weighted by 2D sensitivity profiles of the coils in the array (i.e. the plurality of coil sensitivity weighted projections are generated), for example, and taking the Fourier transform along the x direction when a phase encoding gradient is applied yields the phase-modulated projection of the sensitivity-weighted image onto the x-axis (i.e. wherein the plurality of coil sensitivity weighted projections are generated using a discrete Fourier transform of the signal), as indicated above), for example).
Regarding claim 10, claim 9 is incorporated and the combination of Kyriakos, Schlemper, Levin, and James, as a whole, teaches the non-transitory computer-accessible medium for magnetic resonance imaging (Kyriakos, Col. 1 and Schlemper, Par. [0008]), wherein the computing arrangement is further configured to concatenate the plurality of coil sensitivity weighted projections (Schlemper, Par. [102-205]: the neural network 224 may be trained using particular loss functions described next… An MRI system may have one or multiple RF coils configured to detect MR signals in the imaging region of the MR system… the neural network 224 may be trained to suppress RF interference ic. To this end, training data may be created that includes all of the components of sc separately so that ground truth is available… input to the neural network 224 may be: (1) the signal sc for each coil, so that the neural network suppresses RF interference for each coil separately; (2) the signals sc for all the coils as separate channels, so that the neural network suppresses RF interference for all coils at the same time; or (3) the signals sc for each coil, as separate channels… Neural network 232 may be used in embodiments in which the MRI system collects data using multiple RF coils… the neural network 232 may be used to combine the images (from among initial images 232) generated from data collected by different RF coils, but corresponding to the same slices. As described in more detail below in the “Coil Estimation” Section below, neural network 232 may be used to either estimate such a combined image directly or to estimate sensitivity profiles for the different RF coils, which in turn may be used to combine the images… the process of synthesizing spatial frequency data from MR image data for training a neural network may take into account one or more characteristics of MRI system that will collect patient data that the neural network is being trained to process once the neural network is deployed. Non-limiting, examples of such characteristics include, but are not limited to… sensitivity of RF coils in the MRI system… a process 500 for generating training data from MR images for training the neural network models described herein, in accordance with some embodiments of the technology described herein. The process 500 starts with a magnitude MR volume 502 using various specified characteristics of an MRI system generates spatial frequency data 550, which includes spatial frequency data collected multiple times (Navg times in this example) by each of multiple RF cols of the MRI system… process 500 may be repeated multiple times by starting from the same MR volume 502 to generate different spatial frequency data 550, since multiple portions of the process 500 can be made to vary across different runs since these portions sample certain variations and parameters at random… at acts 526 and 528, an RF coil sensitivity profile is generated for each of the Ncoil RF coils to obtain multiple RF coil sensitivity profiles… Each generated RF coil sensitivity profile Si is complex-valued, with the magnitudes generated at act 526 using one or more RF coil models and with the phases generated (e.g., randomly) at 528. The resulting RF sensitivity profiles are applied to the MR volume (e.g., to the result of performing, at 524, pulse sequence specific augmentation on target MR volume 520) to obtain multiple MR volumes, each of the multiple MR volumes obtained by applying a respective RF coil sensitivity profile to the MR volume resulting at the output of 524… The RF coil model used at 524 may be of any suitable type… the RF coil model used at 526 may be a physics-based RF coil model, which may be configured to calculate the sensitivity of a particular RF coil given its geometry… a non-uniform Fourier transform is applied to the noise-corrupted coil-weighted MR volumes produced at 542; Par. [0346-386]: an ensembled MR image may be generated from the plurality of MR images… Combining the plurality of transformed MR images to obtain the ensembled MR image may comprise, for example, performing an average or a weighted average (e.g., adding images weighted by positive and/or negative weights)… generating the MR image of the subject using the plurality of MR images and the plurality of RF coil profiles comprises generating the MR image of the subject as a weighted combination of the plurality of MR images, each of the plurality of MR images being weighted by a respective RF coil profile in the plurality of RF coil profiles. In some embodiments, the plurality of MR images comprises a first MR image generated from a first input MR dataset obtained using a first RF coil of the plurality of RF coils, and wherein generating the MR image of the subject comprises weighting different pixels of the first MR image using different values of a first RF coil profile among the plurality of RF coil profiles, the first RF coil profile being associated with the first RF coil… training the neural network may comprise generating training data by simulating complex phase for various MR images and training the neural network to predict the coil profile from complex-valued image data. In some embodiments, the neural network may take as input individual coil reconstructions and produce the corresponding estimated coil profile… or take all Ncoil input and produce Ncoil sensitivity profiles jointly. Given the dataset D that contains the coil weighted images… and the ground truth sensitivity maps; wherein the computing arrangement is further configured to concatenate the plurality of coil sensitivity weighted projections (e.g. system and method for generating magnetic resonance (MR) images of a subject from MR data obtained by a magnetic resonance imaging (MRI) system, in which a plurality of images is generated from a portion of input MR spatial frequency data collected by a respective RF coil in a plurality of RF coils of the MRI system, for example, including at least one processor configured to perform functions (i.e. the computing arrangement is further configured to), and computing a two-dimensional sensitivity profile for each receiver coil in a multi-dimensional array that is recorded to the memory device, by using phase modulated projections of received magnetic resonance data onto a frequency encoded x-axis, weighted by 2D sensitivity profiles of the coils in the array (i.e. the plurality of coil sensitivity weighted projections), for example, including a neural network used to either estimate a combined image directly or to estimate sensitivity profiles for different RF coils, which in turn are used to combine (i.e. concatenate) the images (i.e. wherein the computing arrangement is further configured to concatenate the plurality of coil sensitivity weighted projections), as indicated above), for example).
The same motivation to combine above-mentioned teachings applies, as previously indicated in claim 1.
Regarding claim 11, claim 8 is incorporated and the combination of Kyriakos, Schlemper, Levin, and James, as a whole, teaches the non-transitory computer-accessible medium for magnetic resonance imaging (Kyriakos, Col. 1 and Schlemper, Par. [0008]), wherein the inverting of the column in the plurality of coil sensitivity weighted projections comprises inverting coil sensitivities for a particular column for all rows (Kyriakos, Col. 2-3: parallel encoding in MRI is achieved by using the sensitivity profiles of an array of RF receiver coils at least partially surrounding the object of interest… Wi(x,y) represents the 2D sensitivity profile of the ith coil of the array, and ρ(x, y) is an image slice in a selected (x, y) plane… equation represents a projection of the phase modulated image ρ(x,y) onto the x-axis. Further, this signal can be represented in discrete form… apparatus and method of the present invention use phase modulated projections of the received magnetic resonance data onto the frequency encoded (x-) axis, weighted by the 2D sensitivity profiles of the coils in the array, in order to reconstruct ρ(x,y) column by column (i.e., orthogonal to the x-axis)… A two-dimensional sensitivity profile for each receiver coil in the multi-dimensional array is then computed and recorded to the memory device. Thereafter, a plurality of magnetic resonance signals of the object of interest located within the imaging volume is acquired from each receiver coil and recorded to the memory device. An image of the object of interest in a desired plane extending transversely through the imaging volume of the magnetic resonance imaging device then is reconstructed line-by line by the central processor. This reconstruction combines the inverse of the matrix of the sensitivity profiles of each receiver coil and the matrix of the recorded MR data signals together to provide an image that may be displayed or printed; Col. 4: an object of interest is located in the imaging volume of the magnetic resonance imaging device. Magnetic resonance data from the plane of interest is acquired from each of the RF receiver coils and stored in memory… determining and storing in memory the point-by-point time difference of signal reception by each of the receiver coils. This information may be combined with the sensitivity information in the inverse of the matrix of the sensitivity profile so as to provide a phase compensated, load-weighted, point-by-point, inverse, sensitivity profile matrix for each said receiver coil… The receiver coils are adapted to acquire simultaneously a plurality of magnetic resonance signals from an object of interest located within the imaging volume, and to transfer the same to the memory device… the central processor is adapted to reconstruct, and to display on the display device, line-by-line, a two-dimensional image taken in a selected plane extending transversely through the object of interest…This is accomplished by multiplying the inverse of the matrix of sensitivity profiles of the receiver coils and the matrix of data signals acquired by the receiver coils, and displaying the resultant product on the display device; Col. 8-9: sensitivity profile information from a number of receiver coils is used in order to minimize the number of acquisitions needed to estimate and reconstruct ρ(x, y). Taking the Fourier Transform of equation (1) along the x direction when a phase encoding gradient Gg y is applied yields… the phase-modulated projection of the sensitivity-weighted image onto the x-axis… With reference to FIG. 4, the Fourier Transform of the signals received in all three coils respectively is equal to the projection of the phase encoded image weighted by the three profiles W1(i,j), W2(i,j), W3(i,j) respectively onto the x axis… Assume the image to be of size 3×3 as shown in FIG. 4, where each pixel has an intensity value denoted by r(i,j), where i is the column number and j is the row number. The goal is to determine and reconstruct the values of r(i,j) for all the pixels of the image… in order to reconstruct the image [p(m,n)], a matrix pseudoinverse of [Amj(k,y)] is computed for every position mj along the x-axis. This yields a column by column reconstruction; Col. 10-11: Gx and Gy represent the gradients applied in the x and y directions respectively. Then taking the fourier transform with respect to x provides… the phase modulated projection of the image r (x,y) weighted by the sensitivity profile Wk(i,j) onto the x-axis… In order to appropriately account for all of the frequencies included in the image, the choice of the phase modulations used in the inversion matrix should be determined by the frequency content of the sensitivity profile; wherein the inverting of the column in the plurality of coil sensitivity weighted projections comprises inverting coil sensitivities for a particular column for all rows (e.g. system and method for generating magnetic resonance (MR) images of a subject from MR data obtained by a magnetic resonance imaging (MRI) system, in which a plurality of images is generated from a portion of input MR spatial frequency data collected by a respective RF coil in a plurality of RF coils of the MRI system, for example, including at least one processor configured to perform functions (i.e. the computing arrangement is further configured to), and computing a two-dimensional sensitivity profile for each receiver coil in a multi-dimensional array that is recorded to the memory device, by using phase modulated projections of received magnetic resonance data onto a frequency encoded x-axis, weighted by 2D sensitivity profiles of the coils in the array (i.e. the plurality of coil sensitivity weighted projections), for example, and using a matrix pseudoinverse computed for every position along the x-axis (i.e. inverting a column in the plurality of coil sensitivity weighted projections to generate inverted column information), which yields a column by column reconstruction, including an image of the object or subject of interest reconstructed by the programmed central processor, by combining the inverse of the matrix of the sensitivity profiles of each receiver coil and the matrix of the recorded MR data signals together to provide an image (i.e. wherein the inverting of the column in the plurality of coil sensitivity weighted projections comprises inverting coil sensitivities for a particular column for all rows), as indicated above, for example) and the channels q (Schlemper, Par. [102-202]: the neural network 224 may be trained using particular loss functions described next… An MRI system may have one or multiple RF coils configured to detect MR signals in the imaging region of the MR system… the neural network 224 may be trained to suppress RF interference ic. To this end, training data may be created that includes all of the components of sc separately so that ground truth is available… input to the neural network 224 may be: (1) the signal sc for each coil, so that the neural network suppresses RF interference for each coil separately; (2) the signals sc for all the coils as separate channels, so that the neural network suppresses RF interference for all coils at the same time; or (3) the signals sc for each coil, as separate channels… Neural network 232 may be used in embodiments in which the MRI system collects data using multiple RF coils… the neural network 232 may be used to combine the images (from among initial images 232) generated from data collected by different RF coils, but corresponding to the same slices. As described in more detail below in the “Coil Estimation” Section below, neural network 232 may be used to either estimate such a combined image directly or to estimate sensitivity profiles for the different RF coils, which in turn may be used to combine the images… the process of synthesizing spatial frequency data from MR image data for training a neural network may take into account one or more characteristics of MRI system that will collect patient data that the neural network is being trained to process once the neural network is deployed. Non-limiting, examples of such characteristics include, but are not limited to… sensitivity of RF coils in the MRI system… a process 500 for generating training data from MR images for training the neural network models described herein, in accordance with some embodiments of the technology described herein. The process 500 starts with a magnitude MR volume 502 using various specified characteristics of an MRI system generates spatial frequency data 550, which includes spatial frequency data collected multiple times (Navg times in this example) by each of multiple RF cols of the MRI system… process 500 may be repeated multiple times by starting from the same MR volume 502 to generate different spatial frequency data 550, since multiple portions of the process 500 can be made to vary across different runs since these portions sample certain variations and parameters at random… at acts 526 and 528, an RF coil sensitivity profile is generated for each of the Ncoil RF coils to obtain multiple RF coil sensitivity profiles… Each generated RF coil sensitivity profile Si is complex-valued, with the magnitudes generated at act 526 using one or more RF coil models and with the phases generated (e.g., randomly) at 528. The resulting RF sensitivity profiles are applied to the MR volume (e.g., to the result of performing, at 524, pulse sequence specific augmentation on target MR volume 520) to obtain multiple MR volumes, each of the multiple MR volumes obtained by applying a respective RF coil sensitivity profile to the MR volume resulting at the output of 524; Par. [0149-155]: the input is processed by repeated application of two convolutions with 3×3 kernels, each followed by application of a non-linearity (e.g., a ReLU), an average 2×2 pooling operation with stride 2 for downsampling. At each downsampling step the number of feature channels is doubled… In the upsampling path, the data is processed be repeated upsampling of the feature map using an average unpooling step that halves the number of feature channels, a concatenation with the corresponding feature map from the downsampling path, and two 3×3 convolutions, each followed by application of a non-linearity… Inversion of A is more involved. For the (approximately) fully-sampled case, one can consider direct inversion… the inversion is ill-posed, and Eq. 1 should be solved by iterative algorithm; Par. [0285]: After obtaining the first and second input MR data, a first set of one or more MR images and a second set of one or more MR images may be generated from the first input MR data in act 830 and from the second input MR data in act 840, respectively, in accordance with some embodiments of the technology described herein. The first and second sets of MR images may be generated, for example, by reconstructing the first and second input MR data to transform the first and second input MR data from the spatial frequency domain to the image domain. The reconstruction may be performed in any suitable way, including linear and non-linear methods. For example, when the spatial frequency domain data is spaced on a Cartesian grid, the data may be transformed using an inverse 2D Fourier transformation (e.g., using the inverse 2D fast Fourier transform)… when the spatial frequency domain data is under-sampled, the data may be transformed using an inverse non-uniform Fourier transformation, using a neural network model (e.g., reconstruction neural network 212); Par. [0404]: RF coil compression allows for improved training of neural networks because each of the virtual RF channels contains more information than the physical RF channels would have, which makes it easier for the neural network training algorithms to extract information for estimating neural network rates, resulting in faster training (e.g., fewer iterations thereby reducing computational resources required for training) and improved performance; and the channels q (e.g. system and method for generating magnetic resonance (MR) images of a subject from MR data obtained by a magnetic resonance imaging (MRI) system, in which a plurality of images is generated from a portion of input MR spatial frequency data collected by a respective RF coil in a plurality of RF coils of the MRI system, for example, including at least one processor configured to perform functions, including spatial frequency data collected multiple times by each of multiple RF coils of the MRI system and signals sc for each coil, as separate channels (i.e. the channels q), for example, including a neural network used to either estimate a combined image directly or to estimate sensitivity profiles for different RF coils, which in turn are used to combine the images, and when the spatial frequency domain data is under-sampled, the data is transformed using an inverse non-uniform Fourier transformation (i.e. inverting coil sensitivities for the channels q), as indicated above), for example).
The same motivation to combine above-mentioned teachings applies, as previously indicated in claim 1.
Regarding claim 12, claim 1 is incorporated and the combination of Kyriakos, Schlemper, Levin, and James, as a whole, teaches the non-transitory computer-accessible medium for magnetic resonance imaging (Kyriakos, Col. 1 and Schlemper, Par. [0008]), wherein the inverted column information includes line-intensity profiles (Kyriakos, Col. 2-3: parallel encoding in MRI is achieved by using the sensitivity profiles of an array of RF receiver coils at least partially surrounding the object of interest… Wi(x,y) represents the 2D sensitivity profile of the ith coil of the array, and ρ(x, y) is an image slice in a selected (x, y) plane… equation represents a projection of the phase modulated image ρ(x,y) onto the x-axis. Further, this signal can be represented in discrete form… apparatus and method of the present invention use phase modulated projections of the received magnetic resonance data onto the frequency encoded (x-) axis, weighted by the 2D sensitivity profiles of the coils in the array, in order to reconstruct ρ(x,y) column by column (i.e., orthogonal to the x-axis)… A two-dimensional sensitivity profile for each receiver coil in the multi-dimensional array is then computed and recorded to the memory device. Thereafter, a plurality of magnetic resonance signals of the object of interest located within the imaging volume is acquired from each receiver coil and recorded to the memory device. An image of the object of interest in a desired plane extending transversely through the imaging volume of the magnetic resonance imaging device then is reconstructed line-by line by the central processor. This reconstruction combines the inverse of the matrix of the sensitivity profiles of each receiver coil and the matrix of the recorded MR data signals together to provide an image that may be displayed or printed; Col. 4: an object of interest is located in the imaging volume of the magnetic resonance imaging device. Magnetic resonance data from the plane of interest is acquired from each of the RF receiver coils and stored in memory… determining and storing in memory the point-by-point time difference of signal reception by each of the receiver coils. This information may be combined with the sensitivity information in the inverse of the matrix of the sensitivity profile so as to provide a phase compensated, load-weighted, point-by-point, inverse, sensitivity profile matrix for each said receiver coil… The receiver coils are adapted to acquire simultaneously a plurality of magnetic resonance signals from an object of interest located within the imaging volume, and to transfer the same to the memory device… the central processor is adapted to reconstruct, and to display on the display device, line-by-line, a two-dimensional image taken in a selected plane extending transversely through the object of interest…This is accomplished by multiplying the inverse of the matrix of sensitivity profiles of the receiver coils and the matrix of data signals acquired by the receiver coils, and displaying the resultant product on the display device; Col. 8-11: sensitivity profile information from a number of receiver coils is used in order to minimize the number of acquisitions needed to estimate and reconstruct ρ(x, y). Taking the Fourier Transform of equation (1) along the x direction when a phase encoding gradient Gg y is applied yields… the phase-modulated projection of the sensitivity-weighted image onto the x-axis… in order to reconstruct the image [p(m,n)], a matrix pseudoinverse of [Amj(k,y)] is computed for every position mj along the x-axis. This yields a column by column reconstruction… With reference to FIG. 4, the Fourier Transform of the signals received in all three coils respectively is equal to the projection of the phase encoded image weighted by the three profiles W1(i,j), W2(i,j), W3(i,j) respectively onto the x axis… Assume the image to be of size 3×3 as shown in FIG. 4, where each pixel has an intensity value denoted by r(i,j), where i is the column number and j is the row number. The goal is to determine and reconstruct the values of r(i,j) for all the pixels of the image… Further assume that three coils 1, 2 and 3 are arrayed around the image, and are used for image acquisition as shown in FIG. 4. These coils have sensitivity profiles denoted by W1(i,j), W2(i,j), W3(i,j), respectively. (see FIGS. 4A-4C)… The sensitivity profiles of the coils are determined from a homogeneous image by placing a water phantom whose pixel intensity values are equal to I(i,j) into the imaging volume, and acquiring its image using any chosen pulse sequence with the body coil whose sensitivity profile Wb(i,j)=1 for all the pixels in the image. The reconstructed image would have pixel intensity values of rb(i,j)=I(i,j) for all pixels… Next, images are acquired using the 3 coils shown in FIG. 4 with respective sensitivity profiles [W1(i,j)], [W2(i,j)], [W3(i,j)] to get three images of pixel intensity values… Taking the point-by-point ratio of the images acquired using the three coils in the array divided by the RF body coil image yields the sensitivity profiles W1(i,j), W2(i,j), and W3(i,j), respectively… Gx and Gy represent the gradients applied in the x and y directions respectively. Then taking the fourier transform with respect to x provides… the phase modulated projection of the image r (x,y) weighted by the sensitivity profile Wk(i,j) onto the x-axis… In order to appropriately account for all of the frequencies included in the image, the choice of the phase modulations used in the inversion matrix should be determined by the frequency content of the sensitivity profile; wherein the inverted column information includes line-intensity profiles (e.g. apparatus and methods for magnetic resonance imaging (MRI) includes reconstructing an image of an object or subject of interest to be examined, such as a patient, for example, by using an MRI system which includes a plurality of gradient coils that produce spatially encoded gradients imposed upon a static magnetic field within a region in which the object or subject to be examined is placed, for example, and computing a two-dimensional sensitivity profile for each receiver coil in a multi-dimensional array that is recorded to the memory device, by using phase modulated projections of received magnetic resonance data onto a frequency encoded x-axis, weighted by 2D sensitivity profiles of the coils in the array, and using a matrix pseudoinverse computed for every position along the x-axis (i.e. the inverted column information), for example, including an image of the object or subject of interest reconstructed by the programmed central processor, where each pixel of the image has an intensity value denoted by r(i,j), where i is the column number and j is the row number (i.e. wherein the inverted column information includes line-intensity profiles), for example, and combining the inverse of the matrix of the sensitivity profiles of each receiver coil and the matrix of the recorded MR data signals together to provide an image that is displayed or printed, as indicated above), for example).
Claims 27-29 and 32-34 are rejected under 35 U.S.C. 103 as being unpatentable over Kyriakos, in view of Levin, in further view of James, as applied to claim 25, and in further view of Schlemper.
Regarding claim 27, claim 25 is incorporated and the combination of Kyriakos, Levin, and James, as a whole, teaches the system (Kyriakos, Col. 1), but fails to teach the following as further recited in claim 27.
However, Schlemper teaches further comprising deblurring the at least one portion of the at least one image using at least one deep learning procedure (Schlemper, Par. [0007-9]: a system comprising at least one processor configured to perform: obtaining input MR spatial frequency data obtained by imaging the subject using the MRI system; generating an MR image of the subject from the input MR spatial frequency data using a neural network model comprising: a pre-reconstruction neural network configured to process the input MR spatial frequency data; a reconstruction neural network configured to generate at least one initial image of the subject from output of the pre-reconstruction neural network; and a post-reconstruction neural network configured to generate the MR image of the subject from the at least one initial image of the subject… at least one non-transitory computer readable storage medium storing processor-executable instructions that, when executed by at least one processor, cause the at least one processor to perform a method for generating magnetic resonance (MR) images of a subject from MR data obtained by a magnetic resonance imaging (MRI) system. The method comprises: obtaining input MR spatial frequency data obtained by imaging the subject using the MRI system; generating an MR image of the subject from the input MR spatial frequency data using a neural network model comprising: a pre-reconstruction neural network configured to process the input MR spatial frequency data; a reconstruction neural network configured to generate at least one initial image of the subject from output of the pre-reconstruction neural network; and a post-reconstruction neural network configured to generate the MR image of the subject from the at least one initial image of the subject… a method for generating magnetic resonance (MR) images of a subject from MR data obtained by a magnetic resonance imaging (MRI) system. The method comprising: obtaining first input MR data obtained by imaging the subject using the MRI system; obtaining second input MR data obtained by imaging the subject using the MRI system; generating a first set of one or more MR images from the first input MR data; generating a second set of one or more MR images from the second input MR data; aligning the first set of MR images and the second set of MR images using a neural network model to obtain aligned first and second sets of MR images, the neural network model comprising a first neural network and a second neural network, the aligning comprising: estimating, using the first neural network, a first transformation between the first set of MR images and the second set of MR images; generating a first updated set of MR images from the second set of MR images using the first transformation; estimating, using the second neural network, a second transformation between the first set of MR images and the first updated set of MR images; and aligning the first set of MR images and the second set of MR images at least in part by using the first transformation and the second transformation; combining the aligned first and second sets of MR images to obtain a combined set of one or more MR images; and outputting the combined set of one or more MR images; Par. [0078-80]: the reconstruction neural network is configured to perform data consistency processing using a non-uniform Fourier transformation for transforming image data to spatial frequency data… the MRI system comprises a plurality of RF coils, the at least one initial image of the subject comprises a plurality of images, each of the plurality of images generated from a portion of the input MR spatial frequency data collected by a respective RF coil in a plurality of RF coils, and the post-reconstruction neural network comprises a first neural network (e.g., neural network 232) configured to estimate a plurality of RF coil profiles corresponding to the plurality of RF coils… the method further comprises: generating the MR image of the subject using the plurality of MR images and the plurality of RF coil profiles… the at least one initial image of the subject comprises a first set of one or more MR images and a second set of one or more MR images, and the post-reconstruction neural network comprises a second neural network (e.g., neural network 234) for aligning the first set of MR images and the second set of MR images; Par. [0117-123]: Neural network 232 may be used in embodiments in which the MRI system collects data using multiple RF coils… the neural network 232 may be used to combine the images (from among initial images 232) generated from data collected by different RF coils, but corresponding to the same slices…neural network 232 may be used to either estimate such a combined image directly or to estimate sensitivity profiles for the different RF coils, which in turn may be used to combine the images… neural network 234 may be configured to align two sets of one or more MR images to each other. In some instances, each set of MR images may correspond to a set of images for a given volume (e.g., a number of 2D slices that may be stacked to constitute a volume). Such an alignment allows for the sets of MR images to be averaged to increase the SNR. Performing the averaging without first performing alignment would introduce blurring due to, for example, movement of the patient during acquisition of the data being averaged… the neural network 236 may be applied after neural network 234 is used to align corresponding sets of images so that blurring is not introduced through the combination performed by neural network 236; Par. [0219-220]: inventors have developed deep learning techniques for aligning sets of images obtained by multiple acquisitions of the same slice and/or volume… the deep learning techniques involve using a cascade of two or more neural networks configured to estimate a transformation (e.g., a non-rigid, an affine, a rigid transformation) between two sets of MR images (each set having one or multiple MR images), and aligning the two sets of images using the estimated transformation. In turn, the two sets of images may be averaged to obtain a combined set of images having a higher SNR than the sets of images themselves… the estimated transformation may indicate one or more rotations and/or translations to align the two sets of images. In some embodiments, the deep learning techniques described herein may be used as part of neural network 234 part of post-reconstruction neural network 214, as described herein including in connection with FIG. 2C; further comprising deblurring the at least one portion of the at least one image using at least one deep learning procedure (e.g. system and method for generating magnetic resonance (MR) images of a subject from MR data obtained by a magnetic resonance imaging (MRI) system, in which a plurality of images is generated from a portion of input MR spatial frequency data collected by a respective RF coil in a plurality of RF coils of the MRI system (i.e. at least one portion of the at least one image), for example, including at least one processor configured to implement a neural network used to align corresponding sets of images, including the plurality of images generated from a portion of the input MR spatial frequency data collected, so that blurring is not introduced (i.e. deblur, prevent blurring, etc.) through the combination performed by the neural network used to align the corresponding sets of images, for example, including deep learning techniques for aligning sets of images obtained by multiple acquisitions of the same slice and/or volume (i.e. deblurring the at least one portion of the at least one image using at least one deep learning procedure), as indicated above), for example).
Kyriakos, Levin, James, and Schlemper are considered to be analogous art because they pertain to MR image processing applications. Therefore, the combined teachings of Kyriakos, Levin, James, and Schlemper, as a whole, would have rendered obvious the invention recited in claim 27 with a reasonable expectation of success in order to modify the non-transitory computer-accessible medium for magnetic resonance imaging including reconstructing of an image of an object or subject of interest to be examined (as disclosed by Kyriakos and Schlemper) with further comprising deblurring the at least one portion of the at least one image using at least one deep learning procedure (as taught by Schlemper, Abstract, Par. [0007-9, 78-80, 117-123, 219-220]) to generate magnetic resonance (MR) images of a subject from MR data obtained by a magnetic resonance imaging (MRI) system and to improve the quality of image reconstruction (Schlemper, Abstract, Par. [0003-8, 189, 233, 313, 404, 419, 421]).
Regarding claim 28, claim 27 is incorporated and the combination of Kyriakos, Levin, James, and Schlemper, as a whole, teaches the method (Kyriakos, Col. 1), further comprising:
receiving a reference scan of at least one part of the at least one patient; and
training the at least one deep learning procedure based on the reference scan (Schlemper, Par. [0186-216]: neural network models described herein may be trained using any suitable neural network training algorithm(s), as aspects of the technology described herein are not limited in this respect… the neural network models described herein may be trained by using one or more iterative optimization techniques to estimate neural network parameters from training data… training data for training a neural network may be generated synthetically from available MR images… magnitude MR images (phase information is typically discarded) may be used to generate corresponding spatial frequency data and the resulting (spatial frequency data, MR image) pairs may be used to train a neural network model, including any of the neural network models described herein… Using characteristics of the MRI system that will collect patient data to generate training data allows for the neural network to learn these characteristics and use them to improve its performance on tasks in the reconstruction pipeline. Moreover, this approach allows the trained neural network models to reconstruct MR images of comparably high quality based on sensor data acquired using MRI hardware and software… a process 500 for generating training data from MR images for training the neural network models described herein… The process 500 starts with a magnitude MR volume 502 using various specified characteristics of an MRI system generates spatial frequency data 550, which includes spatial frequency data collected multiple times… by each of multiple RF colis of the MRI system… Process 500 may be performed by any suitable computing device(s)… process 500 begins by accessing a reference magnitude MR volume 502. The MR volume 502 may comprise one or multiple images. Each of the image(s) may represent an anatomical slice of a subject being imaged. The MR volume 502 may include one or more magnitude images obtained by a clinical MRI system… neural network models developed by the inventors and described herein may be trained using training data generated from existing high-field image data… a training dataset of (sensor input data, image) pairs may be generated by, for each pair, starting with a high-field source image xh … training the neural network with data pairs derived from high-field data (as above), but also augmenting the loss function with losses computed with respect to available low-field images. The key insight is that, even if a neural network were trained using high-field data, the resulting network should reconstruct the same image from both: (1) a first set of low-field k-space data; and (2) a second set of low-field data obtained by applying a geometric transformation to the first set of low-field k-space data, where the image reconstruction should be invariant under the transformation… another way to generate a training dataset is to use source images of higher quality xo, such as those obtained from low-field scanners, but using more data samples. The sensor data can be obtained directly by collecting the scanner measurements yo; Par. [0303]: additional aspects of training neural networks configured to perform motion estimation and/or correction are described… It may, in some instances, be difficult to acquire large scale real motion-corrupted data for training of any of the neural network models described herein. Accordingly… it may be desirable to generate synthetic training data including reference MR images and synthetic motion-corrupted MR images based on existing datasets 1002 of MR images; Par. [0389]: training data for training a neural network for estimating coil profiles may be generated synthetically from a dataset of existing MR scans. For example… an MR image x may be loaded from a dataset and random phase may be added to this image to obtain a complex-valued image (since only magnitudes are typically available in existing datasets). Complex-valued coil profiles Si for Ncoil coils may be synthesized next. For example, the sensitivity values for particular pixels/voxels may be sampled according to a Gaussian distribution and random phase may be added; further comprising:
receiving a reference scan of at least one part of the at least one patient; and
training the at least one deep learning procedure based on the reference scan (e.g. system and method for generating magnetic resonance (MR) images of a subject (i.e. at least one patient) from MR data obtained by a magnetic resonance imaging (MRI) system, in which a plurality of images is generated from a portion of input MR spatial frequency data collected by a respective RF coil in a plurality of RF coils of the MRI system (i.e. at least one part of the at least one patient), for example, including at least one processor configured to implement a neural network using deep learning techniques (i.e. at least one deep learning procedure), including obtaining (i.e. receiving) training data for training a neural network (i.e. training the at least one deep learning procedure) for estimating coil profiles, which is generated synthetically from a dataset of existing (i.e. reference, source, etc.) MR scans (i.e. receiving a reference scan of at least one part of the at least one patient), for example, by using characteristics of the MRI system that will collect subject or patient data to generate training data (i.e. receiving a reference scan of at least one part of the at least one patient; and training the at least one deep learning procedure based on the reference scan), as indicated above), for example).
The same motivation to combine above-mentioned teachings applies, as previously indicated in claim 27.
Regarding claim 29, claim 28 is incorporated and the combination of Kyriakos, Levin, James, and Schlemper, as a whole, teaches the method (Kyriakos, Col. 1), further comprising:
generating a plurality of training images by varying at least one of (i) an amplitude of the reference scan, or (ii) a noise level of the reference scan; and
training the at least one deep learning procedure based on the plurality of training images (Schlemper, Par. [0186-216]: neural network models described herein may be trained using any suitable neural network training algorithm(s), as aspects of the technology described herein are not limited in this respect… the neural network models described herein may be trained by using one or more iterative optimization techniques to estimate neural network parameters from training data… training data for training a neural network may be generated synthetically from available MR images… magnitude MR images (phase information is typically discarded) may be used to generate corresponding spatial frequency data and the resulting (spatial frequency data, MR image) pairs may be used to train a neural network model, including any of the neural network models described herein… Using characteristics of the MRI system that will collect patient data to generate training data allows for the neural network to learn these characteristics and use them to improve its performance on tasks in the reconstruction pipeline. Moreover, this approach allows the trained neural network models to reconstruct MR images of comparably high quality based on sensor data acquired using MRI hardware and software… a process 500 for generating training data from MR images for training the neural network models described herein… The process 500 starts with a magnitude MR volume 502 using various specified characteristics of an MRI system generates spatial frequency data 550, which includes spatial frequency data collected multiple times… by each of multiple RF colis of the MRI system… Process 500 may be performed by any suitable computing device(s)… process 500 begins by accessing a reference magnitude MR volume 502. The MR volume 502 may comprise one or multiple images. Each of the image(s) may represent an anatomical slice of a subject being imaged. The MR volume 502 may include one or more magnitude images obtained by a clinical MRI system… an RF coil sensitivity profile is generated for each of the Ncoil RF coils to obtain multiple RF coil sensitivity profiles… Each generated RF coil sensitivity profile Si is complex-valued, with the magnitudes generated at act 526 using one or more RF coil models and with the phases generated (e.g., randomly) at 528. The resulting RF sensitivity profiles are applied to the MR volume (e.g., to the result of performing, at 524, pulse sequence specific augmentation on target MR volume 520) to obtain multiple MR volumes, each of the multiple MR volumes obtained by applying a respective RF coil sensitivity profile to the MR volume resulting at the output of 524… at 538 and 540, correlated Gaussian noise is generated and added, at 542, to the multiple MR volumes produced at 536… the Gaussian noise may be generated by: (1) determining, at 538, a noise level σi for each of the coils; and (2) generating, at 540, Gaussian noise having the covariance of LDLT, where D is a diagonal matrix… and L is the coil correlation matrix determined at 534… at 544, a k-space sampling trajectory is selected… Next, at 546, noise δk(t) is added to sampling trajectory k(t). The noise may be added to simulate for various MRI system imperfections and/or any other reason. Next, at 548, a non-uniform Fourier transform is applied to the noise-corrupted coil-weighted MR volumes produced at 542… The resulting spatial frequency data are then output, at 550. These data may be used for training any of the neural network models described herein…neural network models developed by the inventors and described herein may be trained using training data generated from existing high-field image data… a training dataset of (sensor input data, image) pairs may be generated by, for each pair, starting with a high-field source image xh … training the neural network with data pairs derived from high-field data (as above), but also augmenting the loss function with losses computed with respect to available low-field images. The key insight is that, even if a neural network were trained using high-field data, the resulting network should reconstruct the same image from both: (1) a first set of low-field k-space data; and (2) a second set of low-field data obtained by applying a geometric transformation to the first set of low-field k-space data, where the image reconstruction should be invariant under the transformation… another way to generate a training dataset is to use source images of higher quality xo, such as those obtained from low-field scanners, but using more data samples. The sensor data can be obtained directly by collecting the scanner measurements yo; Par. [0303-308]: additional aspects of training neural networks configured to perform motion estimation and/or correction are described… It may, in some instances, be difficult to acquire large scale real motion-corrupted data for training of any of the neural network models described herein. Accordingly… it may be desirable to generate synthetic training data including reference MR images and synthetic motion-corrupted MR images based on existing datasets 1002 of MR images… To generate such synthetic training datasets, a volume may be selected and loaded in act 1004 from dataset 1002… only a magnitude portion of the volume may be loaded. After loading the selected volume in act 1004, a random affine transformation matrix T may be sampled in act 1006… The transformed volume may be stored as a reference volume… To better train the neural network model, it may be desirable to include synthetic noise in the synthetic training data (e.g., to simulate non-ideal MR imaging conditions). In act 1014, Gaussian noise may be sampled in act 1014. The Gaussian noise may be selected to match the volume size of the loaded volume… noise may be added to the reference volume and the moving volume by undersampling a percentage of the MR data in k-space. In act 1016, the Gaussian noise may be added to the reference volume and the moving volume to form the synthetic training data pair for use by the neural network model; Par. [0389]: training data for training a neural network for estimating coil profiles may be generated synthetically from a dataset of existing MR scans. For example… an MR image x may be loaded from a dataset and random phase may be added to this image to obtain a complex-valued image (since only magnitudes are typically available in existing datasets). Complex-valued coil profiles Si for Ncoil coils may be synthesized next. For example, the sensitivity values for particular pixels/voxels may be sampled according to a Gaussian distribution and random phase may be added; further comprising:
generating a plurality of training images by varying at least one of (i) an amplitude of the reference scan, or (ii) a noise level of the reference scan; and
training the at least one deep learning procedure based on the plurality of training images (e.g. system and method for generating magnetic resonance (MR) images of a subject (i.e. at least one patient) from MR data obtained by a magnetic resonance imaging (MRI) system, in which a plurality of images is generated from a portion of input MR spatial frequency data collected by a respective RF coil in a plurality of RF coils of the MRI system, for example, including at least one processor configured to implement a neural network using deep learning techniques (i.e. at least one deep learning procedure), including obtaining (i.e. receiving) training data for training a neural network for estimating coil profiles, which is generated synthetically from a dataset of existing (i.e. reference, source, etc.) MR scans (i.e. MR images), for example, by using characteristics of the MRI system that will collect subject or patient data to generate training data (i.e. generating a plurality of training images), including (i.e. at least one of) adding noise to a reference volume (i.e. varying at least one of a noise level of the reference scan) to simulate for various MRI system imperfections, and/or any other reason, for example, in order to form the synthetic training data for use by the neural network during training (i.e. and training the at least one deep learning procedure based on the plurality of training images), as indicated above), for example).
The same motivation to combine above-mentioned teachings applies, as previously indicated in claim 27.
Regarding claim 32, claim 30 is incorporated and the combination of Kyriakos, Levin, James, and Schlemper, as a whole, teaches the method (Kyriakos, Col. 1), wherein the signal includes a coil sensitivity for each location of each of the channels q (Schlemper, Par. [102-202]: the neural network 224 may be trained using particular loss functions described next… An MRI system may have one or multiple RF coils configured to detect MR signals in the imaging region of the MR system… the neural network 224 may be trained to suppress RF interference ic. To this end, training data may be created that includes all of the components of sc separately so that ground truth is available… input to the neural network 224 may be: (1) the signal sc for each coil, so that the neural network suppresses RF interference for each coil separately; (2) the signals sc for all the coils as separate channels, so that the neural network suppresses RF interference for all coils at the same time; or (3) the signals sc for each coil, as separate channels… Neural network 232 may be used in embodiments in which the MRI system collects data using multiple RF coils… the neural network 232 may be used to combine the images (from among initial images 232) generated from data collected by different RF coils, but corresponding to the same slices. As described in more detail below in the “Coil Estimation” Section below, neural network 232 may be used to either estimate such a combined image directly or to estimate sensitivity profiles for the different RF coils, which in turn may be used to combine the images… the process of synthesizing spatial frequency data from MR image data for training a neural network may take into account one or more characteristics of MRI system that will collect patient data that the neural network is being trained to process once the neural network is deployed. Non-limiting, examples of such characteristics include, but are not limited to… sensitivity of RF coils in the MRI system… a process 500 for generating training data from MR images for training the neural network models described herein, in accordance with some embodiments of the technology described herein. The process 500 starts with a magnitude MR volume 502 using various specified characteristics of an MRI system generates spatial frequency data 550, which includes spatial frequency data collected multiple times (Navg times in this example) by each of multiple RF cols of the MRI system… process 500 may be repeated multiple times by starting from the same MR volume 502 to generate different spatial frequency data 550, since multiple portions of the process 500 can be made to vary across different runs since these portions sample certain variations and parameters at random… at acts 526 and 528, an RF coil sensitivity profile is generated for each of the Ncoil RF coils to obtain multiple RF coil sensitivity profiles… Each generated RF coil sensitivity profile Si is complex-valued, with the magnitudes generated at act 526 using one or more RF coil models and with the phases generated (e.g., randomly) at 528. The resulting RF sensitivity profiles are applied to the MR volume (e.g., to the result of performing, at 524, pulse sequence specific augmentation on target MR volume 520) to obtain multiple MR volumes, each of the multiple MR volumes obtained by applying a respective RF coil sensitivity profile to the MR volume resulting at the output of 524… The RF coil model used at 524 may be of any suitable type… the RF coil model used at 526 may be a physics-based RF coil model, which may be configured to calculate the sensitivity of a particular RF coil given its geometry; wherein the signal includes a coil sensitivity for each location of each of the channels q (e.g. system and method for generating magnetic resonance (MR) images of a subject from MR data obtained by a magnetic resonance imaging (MRI) system, in which a plurality of images is generated from a portion of input MR spatial frequency data collected by a respective RF coil in a plurality of RF coils of the MRI system, for example, including at least one processor configured to perform functions, including spatial frequency data collected multiple times (Navg times in this example) by each of multiple RF coils of the MRI system and signals sc for each coil, as separate channels (i.e. each location of each of the channels q), for example, including a neural network used to either estimate a combined image directly or to estimate sensitivity profiles for different RF coils, which in turn are used to combine the images (i.e. wherein the signal includes a coil sensitivity for each location of each of the channels q), as indicated above), for example).
The same motivation to combine above-mentioned teachings applies, as previously indicated in claim 27.
Regarding claim 33, claim 32 is incorporated and the combination of Kyriakos, Levin, James, and Schlemper, as a whole, teaches the method (Kyriakos, Col. 1), wherein the plurality of coil sensitivity weighted projections are generated using a discrete Fourier transform of the signal, or wherein the inverting of the column in the plurality of coil sensitivity weighted projections comprises inverting coil sensitivities for a particular column for all rows and the channels q (Kyriakos, Col. 2-3: parallel encoding in MRI is achieved by using the sensitivity profiles of an array of RF receiver coils at least partially surrounding the object of interest… Wi(x,y) represents the 2D sensitivity profile of the ith coil of the array, and ρ(x, y) is an image slice in a selected (x, y) plane… equation represents a projection of the phase modulated image ρ(x,y) onto the x-axis. Further, this signal can be represented in discrete form… apparatus and method of the present invention use phase modulated projections of the received magnetic resonance data onto the frequency encoded (x-) axis, weighted by the 2D sensitivity profiles of the coils in the array, in order to reconstruct ρ(x,y) column by column (i.e., orthogonal to the x-axis)… A two-dimensional sensitivity profile for each receiver coil in the multi-dimensional array is then computed and recorded to the memory device. Thereafter, a plurality of magnetic resonance signals of the object of interest located within the imaging volume is acquired from each receiver coil and recorded to the memory device. An image of the object of interest in a desired plane extending transversely through the imaging volume of the magnetic resonance imaging device then is reconstructed line-by line by the central processor. This reconstruction combines the inverse of the matrix of the sensitivity profiles of each receiver coil and the matrix of the recorded MR data signals together to provide an image that may be displayed or printed; Col. 4: an object of interest is located in the imaging volume of the magnetic resonance imaging device. Magnetic resonance data from the plane of interest is acquired from each of the RF receiver coils and stored in memory… determining and storing in memory the point-by-point time difference of signal reception by each of the receiver coils. This information may be combined with the sensitivity information in the inverse of the matrix of the sensitivity profile so as to provide a phase compensated, load-weighted, point-by-point, inverse, sensitivity profile matrix for each said receiver coil… The receiver coils are adapted to acquire simultaneously a plurality of magnetic resonance signals from an object of interest located within the imaging volume, and to transfer the same to the memory device… the central processor is adapted to reconstruct, and to display on the display device, line-by-line, a two-dimensional image taken in a selected plane extending transversely through the object of interest…This is accomplished by multiplying the inverse of the matrix of sensitivity profiles of the receiver coils and the matrix of data signals acquired by the receiver coils, and displaying the resultant product on the display device; Col. 8-9: sensitivity profile information from a number of receiver coils is used in order to minimize the number of acquisitions needed to estimate and reconstruct ρ(x, y). Taking the Fourier Transform of equation (1) along the x direction when a phase encoding gradient Gg y is applied yields… the phase-modulated projection of the sensitivity-weighted image onto the x-axis… With reference to FIG. 4, the Fourier Transform of the signals received in all three coils respectively is equal to the projection of the phase encoded image weighted by the three profiles W1(i,j), W2(i,j), W3(i,j) respectively onto the x axis… Assume the image to be of size 3×3 as shown in FIG. 4, where each pixel has an intensity value denoted by r(i,j), where i is the column number and j is the row number. The goal is to determine and reconstruct the values of r(i,j) for all the pixels of the image… in order to reconstruct the image [p(m,n)], a matrix pseudoinverse of [Amj(k,y)] is computed for every position mj along the x-axis. This yields a column by column reconstruction; Col. 10-11: Gx and Gy represent the gradients applied in the x and y directions respectively. Then taking the fourier transform with respect to x provides… the phase modulated projection of the image r (x,y) weighted by the sensitivity profile Wk(i,j) onto the x-axis… In order to appropriately account for all of the frequencies included in the image, the choice of the phase modulations used in the inversion matrix should be determined by the frequency content of the sensitivity profile; wherein the plurality of coil sensitivity weighted projections are generated using a discrete Fourier transform of the signal, or wherein the inverting of the column in the plurality of coil sensitivity weighted projections comprises inverting coil sensitivities for a particular column for all rows (e.g. system and method for generating magnetic resonance (MR) images of a subject from MR data obtained by a magnetic resonance imaging (MRI) system, in which a plurality of images is generated from a portion of input MR spatial frequency data collected by a respective RF coil in a plurality of RF coils of the MRI system, for example, including at least one processor configured to perform functions, and computing a two-dimensional sensitivity profile for each receiver coil in a multi-dimensional array that is recorded to the memory device, by using phase modulated projections of received magnetic resonance data onto a frequency encoded x-axis, weighted by 2D sensitivity profiles of the coils in the array (i.e. the plurality of coil sensitivity weighted projections), for example, and taking the Fourier transform along the x direction when a phase encoding gradient is applied yields the phase-modulated projection of the sensitivity-weighted image onto the x-axis (i.e. wherein the plurality of coil sensitivity weighted projections are generated using a discrete Fourier transform of the signal), for example, and using a matrix pseudoinverse computed for every position along the x-axis (i.e. inverting a column in the plurality of coil sensitivity weighted projections to generate inverted column information), which yields a column by column reconstruction, including an image of the object or subject of interest reconstructed by the programmed central processor, by combining the inverse of the matrix of the sensitivity profiles of each receiver coil and the matrix of the recorded MR data signals together to provide an image (i.e. wherein the inverting of the column in the plurality of coil sensitivity weighted projections comprises inverting coil sensitivities for a particular column for all rows), as indicated above, for example) and the channels q (Schlemper, Par. [102-202]: the neural network 224 may be trained using particular loss functions described next… An MRI system may have one or multiple RF coils configured to detect MR signals in the imaging region of the MR system… the neural network 224 may be trained to suppress RF interference ic. To this end, training data may be created that includes all of the components of sc separately so that ground truth is available… input to the neural network 224 may be: (1) the signal sc for each coil, so that the neural network suppresses RF interference for each coil separately; (2) the signals sc for all the coils as separate channels, so that the neural network suppresses RF interference for all coils at the same time; or (3) the signals sc for each coil, as separate channels… Neural network 232 may be used in embodiments in which the MRI system collects data using multiple RF coils… the neural network 232 may be used to combine the images (from among initial images 232) generated from data collected by different RF coils, but corresponding to the same slices. As described in more detail below in the “Coil Estimation” Section below, neural network 232 may be used to either estimate such a combined image directly or to estimate sensitivity profiles for the different RF coils, which in turn may be used to combine the images… the process of synthesizing spatial frequency data from MR image data for training a neural network may take into account one or more characteristics of MRI system that will collect patient data that the neural network is being trained to process once the neural network is deployed. Non-limiting, examples of such characteristics include, but are not limited to… sensitivity of RF coils in the MRI system… a process 500 for generating training data from MR images for training the neural network models described herein, in accordance with some embodiments of the technology described herein. The process 500 starts with a magnitude MR volume 502 using various specified characteristics of an MRI system generates spatial frequency data 550, which includes spatial frequency data collected multiple times (Navg times in this example) by each of multiple RF cols of the MRI system… process 500 may be repeated multiple times by starting from the same MR volume 502 to generate different spatial frequency data 550, since multiple portions of the process 500 can be made to vary across different runs since these portions sample certain variations and parameters at random… at acts 526 and 528, an RF coil sensitivity profile is generated for each of the Ncoil RF coils to obtain multiple RF coil sensitivity profiles… Each generated RF coil sensitivity profile Si is complex-valued, with the magnitudes generated at act 526 using one or more RF coil models and with the phases generated (e.g., randomly) at 528. The resulting RF sensitivity profiles are applied to the MR volume (e.g., to the result of performing, at 524, pulse sequence specific augmentation on target MR volume 520) to obtain multiple MR volumes, each of the multiple MR volumes obtained by applying a respective RF coil sensitivity profile to the MR volume resulting at the output of 524; Par. [0149-155]: the input is processed by repeated application of two convolutions with 3×3 kernels, each followed by application of a non-linearity (e.g., a ReLU), an average 2×2 pooling operation with stride 2 for downsampling. At each downsampling step the number of feature channels is doubled… In the upsampling path, the data is processed be repeated upsampling of the feature map using an average unpooling step that halves the number of feature channels, a concatenation with the corresponding feature map from the downsampling path, and two 3×3 convolutions, each followed by application of a non-linearity… Inversion of A is more involved. For the (approximately) fully-sampled case, one can consider direct inversion… the inversion is ill-posed, and Eq. 1 should be solved by iterative algorithm; Par. [0285]: After obtaining the first and second input MR data, a first set of one or more MR images and a second set of one or more MR images may be generated from the first input MR data in act 830 and from the second input MR data in act 840, respectively, in accordance with some embodiments of the technology described herein. The first and second sets of MR images may be generated, for example, by reconstructing the first and second input MR data to transform the first and second input MR data from the spatial frequency domain to the image domain. The reconstruction may be performed in any suitable way, including linear and non-linear methods. For example, when the spatial frequency domain data is spaced on a Cartesian grid, the data may be transformed using an inverse 2D Fourier transformation (e.g., using the inverse 2D fast Fourier transform)… when the spatial frequency domain data is under-sampled, the data may be transformed using an inverse non-uniform Fourier transformation, using a neural network model (e.g., reconstruction neural network 212); Par. [0404]: RF coil compression allows for improved training of neural networks because each of the virtual RF channels contains more information than the physical RF channels would have, which makes it easier for the neural network training algorithms to extract information for estimating neural network rates, resulting in faster training (e.g., fewer iterations thereby reducing computational resources required for training) and improved performance; and the channels q (e.g. system and method for generating magnetic resonance (MR) images of a subject from MR data obtained by a magnetic resonance imaging (MRI) system, in which a plurality of images is generated from a portion of input MR spatial frequency data collected by a respective RF coil in a plurality of RF coils of the MRI system, for example, including at least one processor configured to perform functions, including spatial frequency data collected multiple times by each of multiple RF coils of the MRI system and signals sc for each coil, as separate channels (i.e. the channels q), for example, including a neural network used to either estimate a combined image directly or to estimate sensitivity profiles for different RF coils, which in turn are used to combine the images, and when the spatial frequency domain data is under-sampled, the data is transformed using an inverse non-uniform Fourier transformation (i.e. inverting coil sensitivities for the channels q), as indicated above), for example).
The same motivation to combine above-mentioned teachings applies, as previously indicated in claim 27.
Regarding claim 34, claim 32 is incorporated and the combination of Kyriakos, Levin, James, and Schlemper, as a whole, teaches the method (Kyriakos, Col. 1), further comprising concatenating the plurality of coil sensitivity weighted projections (Schlemper, Par. [102-205]: the neural network 224 may be trained using particular loss functions described next… An MRI system may have one or multiple RF coils configured to detect MR signals in the imaging region of the MR system… the neural network 224 may be trained to suppress RF interference ic. To this end, training data may be created that includes all of the components of sc separately so that ground truth is available… input to the neural network 224 may be: (1) the signal sc for each coil, so that the neural network suppresses RF interference for each coil separately; (2) the signals sc for all the coils as separate channels, so that the neural network suppresses RF interference for all coils at the same time; or (3) the signals sc for each coil, as separate channels… Neural network 232 may be used in embodiments in which the MRI system collects data using multiple RF coils… the neural network 232 may be used to combine the images (from among initial images 232) generated from data collected by different RF coils, but corresponding to the same slices. As described in more detail below in the “Coil Estimation” Section below, neural network 232 may be used to either estimate such a combined image directly or to estimate sensitivity profiles for the different RF coils, which in turn may be used to combine the images… the process of synthesizing spatial frequency data from MR image data for training a neural network may take into account one or more characteristics of MRI system that will collect patient data that the neural network is being trained to process once the neural network is deployed. Non-limiting, examples of such characteristics include, but are not limited to… sensitivity of RF coils in the MRI system… a process 500 for generating training data from MR images for training the neural network models described herein, in accordance with some embodiments of the technology described herein. The process 500 starts with a magnitude MR volume 502 using various specified characteristics of an MRI system generates spatial frequency data 550, which includes spatial frequency data collected multiple times (Navg times in this example) by each of multiple RF cols of the MRI system… process 500 may be repeated multiple times by starting from the same MR volume 502 to generate different spatial frequency data 550, since multiple portions of the process 500 can be made to vary across different runs since these portions sample certain variations and parameters at random… at acts 526 and 528, an RF coil sensitivity profile is generated for each of the Ncoil RF coils to obtain multiple RF coil sensitivity profiles… Each generated RF coil sensitivity profile Si is complex-valued, with the magnitudes generated at act 526 using one or more RF coil models and with the phases generated (e.g., randomly) at 528. The resulting RF sensitivity profiles are applied to the MR volume (e.g., to the result of performing, at 524, pulse sequence specific augmentation on target MR volume 520) to obtain multiple MR volumes, each of the multiple MR volumes obtained by applying a respective RF coil sensitivity profile to the MR volume resulting at the output of 524… The RF coil model used at 524 may be of any suitable type… the RF coil model used at 526 may be a physics-based RF coil model, which may be configured to calculate the sensitivity of a particular RF coil given its geometry… a non-uniform Fourier transform is applied to the noise-corrupted coil-weighted MR volumes produced at 542; Par. [0346-386]: an ensembled MR image may be generated from the plurality of MR images… Combining the plurality of transformed MR images to obtain the ensembled MR image may comprise, for example, performing an average or a weighted average (e.g., adding images weighted by positive and/or negative weights)… generating the MR image of the subject using the plurality of MR images and the plurality of RF coil profiles comprises generating the MR image of the subject as a weighted combination of the plurality of MR images, each of the plurality of MR images being weighted by a respective RF coil profile in the plurality of RF coil profiles. In some embodiments, the plurality of MR images comprises a first MR image generated from a first input MR dataset obtained using a first RF coil of the plurality of RF coils, and wherein generating the MR image of the subject comprises weighting different pixels of the first MR image using different values of a first RF coil profile among the plurality of RF coil profiles, the first RF coil profile being associated with the first RF coil… training the neural network may comprise generating training data by simulating complex phase for various MR images and training the neural network to predict the coil profile from complex-valued image data. In some embodiments, the neural network may take as input individual coil reconstructions and produce the corresponding estimated coil profile… or take all Ncoil input and produce Ncoil sensitivity profiles jointly. Given the dataset D that contains the coil weighted images… and the ground truth sensitivity maps; further comprising concatenating the plurality of coil sensitivity weighted projections (e.g. system and method for generating magnetic resonance (MR) images of a subject from MR data obtained by a magnetic resonance imaging (MRI) system, in which a plurality of images is generated from a portion of input MR spatial frequency data collected by a respective RF coil in a plurality of RF coils of the MRI system, for example, including at least one processor configured to perform functions, and computing a two-dimensional sensitivity profile for each receiver coil in a multi-dimensional array that is recorded to the memory device, by using phase modulated projections of received magnetic resonance data onto a frequency encoded x-axis, weighted by 2D sensitivity profiles of the coils in the array (i.e. the plurality of coil sensitivity weighted projections), for example, including a neural network used to either estimate a combined image directly or to estimate sensitivity profiles for different RF coils, which in turn are used to combine (i.e. concatenate) the images (i.e. concatenating the plurality of coil sensitivity weighted projections), as indicated above), for example).
The same motivation to combine above-mentioned teachings applies, as previously indicated in claim 27.
Conclusion
Applicant’s amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any extension fee pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Contact
Any inquiry concerning this communication or earlier communications from the examiner should be directed to GUILLERMO RIVERA-MARTINEZ whose telephone number is 571-272-4979. The examiner can normally be reached on Monday-Friday (8am - 5pm Eastern Time). If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Andrew Bee can be reached on 571-270-5183. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/GUILLERMO M RIVERA-MARTINEZ/ Primary Examiner, Art Unit 2677
Prosecution Insights