DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Priority
Acknowledgement is made to Applicant’s claim to priority to U.S. Provisional App. No. 63/333,373 filed April 21, 2022.
Status of Claims
This Office Action is responsive to the claims filed on 08/11/2025. Claim 15 has been amended. Claims 1-19 are presently pending in this application.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claim 1-19 are rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception (i.e., an abstract idea) without significantly more.
Subject matter eligibility pursuant to 35 U.S.C. § 101 requires first (“Eligibility Step 1”) that the claimed invention fall within one of the four statutorily authorized categories, and second (“Eligibility Step 2”) that the claim not be improperly directed to a judicial exception. MPEP 2106(III).
Determination as to whether a claim is improperly directed to a judicial exception is a two-part inquiry. Part one (“Eligibility Step 2A”) depends first (“Eligibility Step 2A, Prong One”) on whether the claim recites a judicial exception, and second (“Eligibility Step 2A, Prong Two”) whether the claim contains additional elements sufficient to integrate the judicial exception into a practical application. MPEP 2106(III). If the claim at issue does recite a judicial exception but does contain said such sufficient additional elements, then the claim is not improperly directed to a judicial exception and is not directed to patent ineligible subject matter. See MPEP 2106.04(d).
If the claim at issue does recite a judicial exception and does not contain sufficient additional elements to integrate the judicial exception into a practical application, then assessment must be made as to whether the claim sufficiently furnishes an inventive concept. MPEP 2106.04(d). Part two of the two-part inquiry (“Eligibility Step 2B”) thus looks at any additional claim elements to determine whether “the claim as a whole amounts to significantly more than the judicial exception itself.” MPEP 2106.05(d), (citing Alice Corp. Pty. Ltd. v. CLS Bank Int'l, 573 U.S. 208, 227-218 (2014) (citing Mayo Collaborative Servs. v. Prometheus Labs., Inc., 566 U.S. 66, at 71-72 (1966))). Claims that do amount to “significantly more” are not directed to patent ineligible subject matter under 35 U.S.C. 101; claims that do not amount to “significantly more” are directed to patent ineligible subject matter. MPEP 2106(III).
Independent claims 1 and 15:
With regard to Step 1, the claims 1 and 15 directed to one of the four statutory categories of invention. Specifically claims 1 and 15 are directed to methods.
With regard to Step 2A: Prong 1, claims 1 and 15 recite limitations directed towards:
Claim 1: generating motion-simulated k-space data…
Claim 15: generating motion artifact classification data; and
analyzing the motion artifact classification data… to control operation of the MRI system
Each of the recited steps, as drafted, amount to nothing more than a limitation that can practically be performed in the human mind and/or with the aid of pen/paper. For example:
Claim 1: The limitations with regard to generating motion-simulated k-space data are broadly recited and can be achieved for a simple 2x2 matrix in the mind or using pen and paper by using a simple 2x2 matrix, applying a motion shift, and transforming the matrix to spatial frequency to generate a simulated k-space data.
Claim 15: The limitations regarding generating motion artifact classification data are broadly recited and can be done in the mind or by using pen and paper by looking at set of k-space data, comparing the data to some data with known motion shift information, and assigning a classification based on the data having the known motions shift.
The limitation regarding analyzing the motion artifact classification data is broadly recited and can be done in the mind or by using pen and paper by looking at a motion classification, determining whether the type of motion is acceptable, and deciding to further continue operation of a MRI system or repeat an acquisition.
Therefore, the limitations recite mental-process type abstract ideas. See MPEP 2106.04(a)(2).
With regard to Step 2A: Prong 2, claims 1 and 15 recite additional elements directed towards:
Claim 1: acquiring k-space data with an MRI system;
accessing magnetic resonance images with a computer system;
accessing motion parameters with the computer system;
assembling, by the computer system, a training dataset from the motion-simulated k-space data;
training a neural network on the training dataset using the computer system;
storing the trained neural network with the computer system.
Claim 15: acquiring k-space data from a subject using the MRI system;
accessing a machine learning model with a computer system, wherein the machine learning model has been trained on training data to detect motion artifacts in k-space data;
inputting the k-space data to the machine learning model;
The limitation, however, directed towards acquiring k-space data, accessing magnetic resonance images, accessing motion parameters, assembling training data, training a neural network, storing the trained neural network, accessing a machine learning model, and inputting the k-space data into the machine learning model are recited at a high level of generality. The limitations of acquiring data, accessing data, assembling data, and training the model is interpreted as merely instructions to implement the abstract ideas on the processors and extra-solution activities of mere data gathering needed to perform the abstract idea on the computing system with the MRI system. Such instructions and activities are not sufficient to integrate a judicial exception into a practical application. See MPEP 2106.05(f) and (g). Consequently, the additional elements do not appear, either individually or as a whole, to integrate the judicial exceptions into a practical application.
With regard to Step 2B, as explained above, the additional limitations of the claims comprise no more than instruction to implement the judicial exceptions on a computer/merely use a computer as a tool to perform the judicial exceptions, and extra-solution steps needed to obtain the data needed to perform the judicial exceptions. Therefore, when considered separately and in combination, the additional limitations do not result in the claims, as a whole, amounting to significantly more than the judicial exceptions.
Dependent claims 2, 3, and 14 add further limitations to the type of motion parameters. These limitations are:
the motion parameters comprise three- dimensional motion parameters
the motion parameters comprise both three-dimensional translations and three-dimensional rotations
the motion parameters comprise non-rigid motion parameters
These additional limitations comprise no more than limitations which further limit the type o of data indicating a field of use in which to apply the judicial exception. See MPEP 2106.05(h). Therefore, the additional limitations, when considered separately and in combination, do not integrate the judicial exceptions into a practical application, or result in the claims amounting to significantly more than the judicial exceptions.
Dependent claims 4, 5, and 10 add further limitations to inputs to the forward model. These limitations are:
accessing pulse sequence data indicating a k-space sampling pattern and including inputting the pulse sequence data to the forward model
indicating a segment ordering for phase-encoding lines for two-dimensional slices in a multislice acquisition
accessing coil sensitivity maps and inputting the coil sensitivity maps as an additional input to the forward model
These additional limitations comprise no more than additional instructions to implement the abstract ideas on the processors and indicating a field of use or technological environment in which to apply the judicial exception. See MPEP 2106.05(f) and (h). Therefore, the additional limitations, when considered separately and in combination, do not integrate the judicial exceptions into a practical application, or result in the claims amounting to significantly more than the judicial exceptions.
Dependent claims 6-9 add further limitations to how the motion artifacts are analyzed. These limitations are:
extract features indicative of motion artifacts and storing the extracted featured in the training dataset
computing a cross-correlation between adjacent phase-encoding lines in the motion-simulated k-space data
labeling with different severities of motion artifact based on a magnitude of the cross-correlation
labeling with different severities of motion artifact based on the magnitude of the cross-correlation in a central region of the k-space
These additional limitations comprise no more than additional instructions to implement the abstract ideas on the processors, insignificant extra-solution activity of mere data gathering, and indicating a field of use or technological environment in which to apply the judicial exception. See MPEP 2106.05(f), (g), and (h). Therefore, the additional limitations, when considered separately and in combination, do not integrate the judicial exceptions into a practical application, or result in the claims amounting to significantly more than the judicial exceptions.
Dependent claims 11, 12, 18, and 19 add further limitations to the type of neural network used. These limitations are:
the neural network is a convolutional neural network
the convolutional neural network comprises a ResNet architecture
the machine learning model is a neural network
the neural network comprises a ResNet architecture
These additional limitations comprise no more than limitations what type of neural network is used indicating a field of use or technological environment in which to apply the judicial exception. See 2106.05 (h). Therefore, the additional limitations, when considered separately and in combination, do not integrate the judicial exceptions into a practical application, or result in the claims amounting to significantly more than the judicial exceptions.
Dependent claim 13 adds further limitations to the functionality of the neural network. These limitations are:
the neural network has a plurality of outputs, wherein each of the plurality of outputs corresponds to a different classification of motion artifact severity
These additional limitations comprise no more than additional instructions to implement the abstract ideas on the processors and indicating a field of use or technological environment in which to apply the judicial exception. See MPEP 2106.05(f) and (h). Therefore, the additional limitations, when considered separately and in combination, do not integrate the judicial exceptions into a practical application, or result in the claims amounting to significantly more than the judicial exceptions.
Dependent claims 16 and 17 add further limitations of displaying an alert and controlling operation of the MRI. These limitations are:
displaying an alert to a user when motion artifacts are detected above a threshold value of severity based on the analyzing of the motion artifact classification data
controlling operation of the MRI system by pausing scanning of the subject when motion artifacts are detected above a threshold value of severity based on the analyzing of the motion artifact classification data
These additional limitations comprise no more than well understood and routine activity and additional limitations to perform insignificant extra-solution activity to implement judicial exception. See MPEP 2105.05(d) and (g). Therefore, the additional limitations, when considered separately and in combination, do not integrate the judicial exceptions into a practical application, or result in the claims amounting to significantly more than the judicial exceptions.
Claim Rejections - 35 USC § 102
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
Claims 1, 11, and 13 are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Schuelke (WO 2021228515 A1).
Regarding claim 1, Schuelke teaches a method for training a neural network to detect motion artifacts in k-space data (pg. 13, ln. 31-pg. 14, ln. 8; The image generating neural network is configured for outputting the synthetic magnetic resonance image data in response to receiving the reference magnetic resonance image data as input; Pg. 43, ln. 14-15; The artifact level in the resulting fused image is estimated using a dedicated motion artifact level estimator; Pg. 37, ln. 16-20; Train a contrast-to-contrast network) acquired with a magnetic resonance imaging (MRI) system (Pg. 31, ln. 20-33; magnetic resonance imaging system 302, Fig. 3), the method comprising:
(a) accessing magnetic resonance images (Pg. 30, ln. 10-19; image generating neural network 122 is configured to receive a reference magnetic resonance image and then output a synthetic magnetic resonance image data 128) with a computer system (Pg. 29, ln. 32-Pg. 30, ln. 9; The computational system 106 containing a memory 110, Fig. 1);
(b) accessing motion parameters with the computer system (Pg. 8, ln. 16-24; computational system to receive a motion signal descriptive of motion of the subject; Pg. 43, ln. 13-22; Generation of the associated training dataset was realized based on motion-free volunteer T2w images as well as an artifact simulation pipeline);
(c) generating motion-simulated k-space data (Pg. 42, ln. 32-Pg. 43, ln. 12; the first motion-free scan is then converted to the target contrast, i.e. the contrast of a second scan that is corrupted by motion artifacts; corresponding k-space profiles of the converted first scan. The selection of profiles for replacement depends on the type of scan and the specific k-space acquisition scheme) with the computer system using a forward model (Pg. 44, ln. 4-13; the artifact-corrupted image in Fig. 16 was generated using forward simulation) to convert the magnetic resonance images to k-space data while using the motion parameters (Pg. 43, ln. 26-Pg.44, ln. 3; shows another magnetic resonance image with motion artifacts caused by intentionally corrupting several lines of k-space data) to apply different degrees of motion to the k-space data (Pg. 42, ln. 32-Pg. 43, ln. 12; To reduce the artifact level for this second scan, certain k-space profiles of the second scan are then replaced by the corresponding k-space profiles of the converted first scan; selection of profiles for replacement depends on the type of scan and the specific k-space acquisition scheme);
(d) assembling, by the computer system, a training dataset from the motion-simulated k-space data (Pg. 43, ln. 13-22; Generation of the associated training dataset was realized based on motion-free volunteer T2w images as well as an artifact simulation pipeline; Pg. 42, ln. 22-Pg. 43, ln. 5; Creation of a suitable dataset can be realized in multiple ways… the first motion-free scan is then converted to the target contrast, i.e. the contrast of a second scan that is corrupted by motion artifacts);
(e) training a neural network on the training dataset using the computer system (Pg. 37, ln. 16-Pg. 38, ln. 15; Train a contrast-to-contrast network N (image generating neural network 122)… This network can be trained using a training dataset of pairs of the same image with contrast A and B); and
(f) storing the trained neural network with the computer system (Pg. 30, ln. 10-19; The memory 110 is further shown as containing an image generating neural network 122, Fig. 1).
Regarding claim 11, Schuelke teaches all of the limitations of claim 1 as noted above.
Schuelke further teaches the neural network is a convolutional neural network (Pg. 41, ln. 9-12; For example, they can be fully convolutional networks such as U-net or variants of it; Pg. 42, ln. 6-7; block 122 is a contrast-to-contrast conversion U-Net neural network which is equivalent to the image generating neural network 122).
Regarding claim 13, Schuelke teaches all of the limitations of claim 1 as noted above.
Schuelke further teaches the neural network has a plurality of outputs (Pg. 42, Ln. 1; network that is trained to estimate the motion artifact level in an image), wherein each of the plurality of outputs corresponds to a different classification of motion artifact severity (Pg. 43, ln. 13-22; If the estimated artifact level in the fused image is considerably lower than the estimated artifact level in the original image, the profile(s) are considered to be corrupted by motion; differentiating between having motion artifacts and not having motion artifacts is considered to read on the claimed limitation of different classifications of motion artifact severity as understood in its broadest reasonable interpretation).
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 2-5, 10, 12, and 14 are rejected under 35 U.S.C. 103 as being unpatentable over Schuelke in view of Schlemper (US 20200294287).
Regarding claim 2, Schuelke teaches all of the limitations of claim 1 as noted above.
Schuelke does not explicitly teach the motion parameters comprise three-dimensional motion parameters.
Schlemper, however, teaches in a similar field of endeavor a method for training a neural network (Paragraph [0092]; Aspects of training neural network model 204 are described herein) to detect motion artifacts in k-space data acquired with a magnetic resonance imaging (MRI) system (Paragraph [0071]; deep-learning processing pipeline for processing MRI data to generate MR images of patients… to one another to compensate for patient motion), the method comprising generating motion-simulated k-space data (Paragraph [0210]; a training dataset of training data generated using a model of the “forward process”, Fig. 5A-C; Paragraph [0194]; updated MR volume x′ 512, Fig. 5A; Paragraph [0205]; Next, at 548, a non-uniform Fourier transform is applied to the noise-corrupted coil-weighted MR volumes produced at 542, Fig. 5C) using the motion parameters (Paragraph [0194]-[0196]; transformations T(x), 508, Fig. 5; one or more 2D or 3D affine transformations (e.g., one or more translations, one or more rotations, one or more scalings)) to apply different degrees of motion to the k-space data (Paragraph [0194]-[0198]; application of a histogram augmentation function I(x) (generated at 510)… generating multiple different training examples from a single underlying MR volume; augmentation function I(r) generated at 510 may be used to change the intensity variations in regions of the image to simulate various effects);
wherein the motion parameters comprise three-dimensional motion parameters (Paragraph [0195]; one or more 2D or 3D affine transformations (e.g., one or more translations, one or more rotations, one or more scalings) and/or one or more 2D or 3D non-rigid transformations (e.g., one or more deformations)).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the motion parameters of Schuelke to comprise three-dimensional motion parameters as taught by Schlemper because it would have been a well understood method for generating motion augmented training data that further would have allowed simulating a realistic variation of how different positions and orientations of a patient's anatomy may be positioned within the MRI system, thereby improving the training dataset and overall improve the training process (Schlemper, paragraph [0196]).
Regarding claim 3, together Schuelke and Schlemper teach all of the limitations of claim 2 as noted above.
Schuelke discloses the invention as claimed and discussed above, but fails to explicitly disclose the motion parameters comprise both three-dimensional translations and three-dimensional rotations.
Schlemper, however, further teaches the motion parameters comprise both three-dimensional translations and three-dimensional rotations (Paragraph [0195]; one or more 2D or 3D affine transformations (e.g., one or more translations, one or more rotations, one or more scalings)).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have further modified the motion parameters of Schuelke in view of Schlemper to have comprised both three-dimensional translations and three-dimensional rotations because it would have allowed simulating a realistic variation of how different positions and orientations of a patient's anatomy may be positioned within the MRI system, thereby improving the training dataset and overall improve the training process (Schlemper, paragraph [0196]).
Regarding claim 4, Schuelke teaches all of the limitations of claim 1 as noted above.
Schuelke further teaches accessing pulse sequence data indicating a k-space sampling pattern (Pg. 13, ln. 7-15; choosing a k-space data sampling pattern for the first pulse sequence commands using the synthetic k-space data).
Schuelke does not explicitly teach generating the motion-simulated k-space data includes inputting the pulse sequence data to the forward model such that the magnetic resonance images are resampled to the k-space sampling pattern.
Schlemper, however, teaches in a similar field of endeavor a method for training a neural network (Paragraph [0092]; Aspects of training neural network model 204 are described herein) to detect motion artifacts in k-space data acquired with a magnetic resonance imaging (MRI) system (Paragraph [0071]; deep-learning processing pipeline for processing MRI data to generate MR images of patients… to one another to compensate for patient motion), the method comprising generating motion-simulated k-space data (Paragraph [0210]; a training dataset of training data generated using a model of the “forward process”, Fig. 5A-C; Paragraph [0194]; updated MR volume x′ 512, Fig. 5A; Paragraph [0205]; Next, at 548, a non-uniform Fourier transform is applied to the noise-corrupted coil-weighted MR volumes produced at 542, Fig. 5C) using the motion parameters (Paragraph [0194]-[0196]; transformations T(x), 508, Fig. 5; one or more 2D or 3D affine transformations (e.g., one or more translations, one or more rotations, one or more scalings)) to apply different degrees of motion to the k-space data (Paragraph [0194]-[0198]; application of a histogram augmentation function I(x) (generated at 510)… generating multiple different training examples from a single underlying MR volume; augmentation function I(r) generated at 510 may be used to change the intensity variations in regions of the image to simulate various effects);
wherein generating the motion-simulated k-space data includes inputting the pulse sequence data to the forward model (Paragraphs [0199]-[0202]; 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, Fig. 5B) such that the magnetic resonance images are resampled to the k-space sampling pattern (Paragraph [0199]; to generate multiple sets of spatial frequency data, each set including spatial frequency data for Ncoil RF coils (8 in this example). Within act 525, first sequence specific augmentation is performed at acts 522 and 524).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the method of Schuelke such that generating the motion-simulated k-space data includes inputting the pulse sequence data to the forward model such that the magnetic resonance images are resampled to the k-space sampling pattern as taught by Schlemper. This would have allowed simulating the types of RF artefacts that may be expected to be observed during a particular pulse sequence in the training dataset which would improve the realism of dataset and thereby improve the overall training of the neural network using the simulated data (Schlemper, paragraph [0200]).
Regarding claim 5, together Schuelke and Schlemper teach all of the limitations of claim 4 as noted above.
Schuelke discloses the invention as claimed and discussed above, but fails to explicitly disclose the pulse sequence data indicate a segment ordering for phase-encoding lines for two-dimensional slices in a multislice acquisition.
Schlemper, however, further teaches the pulse sequence data indicate a segment ordering for phase-encoding lines (Paragraph [0198]-[0202]; an RF coil sensitivity profile is generated for each of the Ncoil RF coils… with the magnitudes generated at act 526 using one or more RF coil models and with the phases generated, Fig. 5B steps 528 and 530) for two-dimensional slices (Paragraph [0192]; Each of the image(s) may represent an anatomical slice of a subject being imaged, Fig. 5) in a multislice acquisition (Paragraph [0192]; process 500 begins by accessing a reference magnitude MR volume 502. The MR volume 502 may comprise one or multiple images).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have further modified the method of Schuelke in view of Schlemper to have included the pulse sequence data indicating a segment ordering for phase-encoding lines for two-dimensional slices in a multislice acquisition because it would have allowed simulating the types of RF artefacts that may be expected to be observed during a particular pulse sequence in the training dataset which would improve the realism of dataset and thereby improve the overall training of the neural network using the simulated data (Schlemper, paragraph [0200]).
Regarding claim 10, Schuelke teaches all of the limitations of claim 1 as noted above.
Schuelke does not explicitly teach accessing coil sensitivity maps and inputting the coil sensitivity maps as an additional input to the forward model.
Schlemper, however, teaches in a similar field of endeavor a method for training a neural network (Paragraph [0092]; Aspects of training neural network model 204 are described herein) to detect motion artifacts in k-space data acquired with a magnetic resonance imaging (MRI) system (Paragraph [0071]; deep-learning processing pipeline for processing MRI data to generate MR images of patients… to one another to compensate for patient motion), the method comprising generating motion-simulated k-space data (Paragraph [0210]; a training dataset of training data generated using a model of the “forward process”, Fig. 5A-C; Paragraph [0194]; updated MR volume x′ 512, Fig. 5A; Paragraph [0205]; Next, at 548, a non-uniform Fourier transform is applied to the noise-corrupted coil-weighted MR volumes produced at 542, Fig. 5C) using the motion parameters (Paragraph [0194]-[0196]; transformations T(x), 508, Fig. 5; one or more 2D or 3D affine transformations (e.g., one or more translations, one or more rotations, one or more scalings)) to apply different degrees of motion to the k-space data (Paragraph [0194]-[0198]; application of a histogram augmentation function I(x) (generated at 510)… generating multiple different training examples from a single underlying MR volume; augmentation function I(r) generated at 510 may be used to change the intensity variations in regions of the image to simulate various effects);
wherein generating the motion-simulated k-space data comprises accessing coil sensitivity maps (Paragraph [0201]; 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, Fig. 5B) and inputting the coil sensitivity maps as an additional input to the forward model (Paragraph [0201]; 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, Fig. 5B).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the method of Schuelke to have further comprised accessing coil sensitivity maps and inputting the coil sensitivity maps as an additional input to the forward model as taught by Schlemper. This would have allowed determining and including any RF coil coupling and/or inductance effects which would affect the data acquisition in the dataset, thereby improving the realism of dataset and thereby improve the overall training of the neural network using the simulated data (Schlemper, paragraph [0202]).
Regarding claim 12, Schuelke teaches all of the limitations of claim 11 as noted above.
Schuelke does not explicitly teach the convolutional neural network comprises a ResNet architecture.
Schlemper, however, teaches in a similar field of endeavor a method for training a neural network (Paragraph [0092]; Aspects of training neural network model 204 are described herein) to detect motion artifacts in k-space data acquired with a magnetic resonance imaging (MRI) system (Paragraph [0071]; deep-learning processing pipeline for processing MRI data to generate MR images of patients… to one another to compensate for patient motion), the method comprising generating motion-simulated k-space data (Paragraph [0210]; a training dataset of training data generated using a model of the “forward process”, Fig. 5A-C; Paragraph [0194]; updated MR volume x′ 512, Fig. 5A; Paragraph [0205]; Next, at 548, a non-uniform Fourier transform is applied to the noise-corrupted coil-weighted MR volumes produced at 542, Fig. 5C) using the motion parameters (Paragraph [0194]-[0196]; transformations T(x), 508, Fig. 5; one or more 2D or 3D affine transformations (e.g., one or more translations, one or more rotations, one or more scalings)) to apply different degrees of motion to the k-space data (Paragraph [0194]-[0198]; application of a histogram augmentation function I(x) (generated at 510)… generating multiple different training examples from a single underlying MR volume; augmentation function I(r) generated at 510 may be used to change the intensity variations in regions of the image to simulate various effects);
wherein the neural network comprises a ResNet architecture (Paragraph [0098]; the neural network 220 may be a convolutional neural network and may have one or more convolutional layers… a ResNet type architecture may be used where convolutional blocks have residual connections).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the neural network of Schuelke to have further comprised a ResNet architecture as taught by Schlemper because it would have been a known architecture for detecting and correcting motion artifacts in MRI data that further would have improved the performance of the neural network when convolutional blocks have residual connections (Schlemper, Paragraph [0098] and [0172]).
Regarding claim 14, Schuelke teaches all of the limitations of claim 1 as noted above.
Schuelke does not explicitly teach the motion parameters comprise non-rigid motion parameters.
Schlemper, however, teaches in a similar field of endeavor a method for training a neural network (Paragraph [0092]; Aspects of training neural network model 204 are described herein) to detect motion artifacts in k-space data acquired with a magnetic resonance imaging (MRI) system (Paragraph [0071]; deep-learning processing pipeline for processing MRI data to generate MR images of patients… to one another to compensate for patient motion), the method comprising generating motion-simulated k-space data (Paragraph [0210]; a training dataset of training data generated using a model of the “forward process”, Fig. 5A-C; Paragraph [0194]; updated MR volume x′ 512, Fig. 5A; Paragraph [0205]; Next, at 548, a non-uniform Fourier transform is applied to the noise-corrupted coil-weighted MR volumes produced at 542, Fig. 5C) using the motion parameters (Paragraph [0194]-[0196]; transformations T(x), 508, Fig. 5; one or more 2D or 3D affine transformations (e.g., one or more translations, one or more rotations, one or more scalings)) to apply different degrees of motion to the k-space data (Paragraph [0194]-[0198]; application of a histogram augmentation function I(x) (generated at 510)… generating multiple different training examples from a single underlying MR volume; augmentation function I(r) generated at 510 may be used to change the intensity variations in regions of the image to simulate various effects);
wherein the motion parameters comprise non-rigid motion parameters (Paragraph [0195]; one or more 2D or 3D non-rigid transformations (e.g., one or more deformations)).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the motion parameters of Schuelke to comprise non-rigid motion parameters as taught by Schlemper because it would have been a well understood method for generating motion augmented training data that further would have allowed simulating a realistic variation of how different positions and orientations of a patient's anatomy may be positioned within the MRI system, thereby improving the training dataset and overall improve the training process (Schlemper, paragraph [0196]).
Claim 6 is rejected under 35 U.S.C. 103 as being unpatentable over Schuelke in view of Mysore Siddu (US 20210279891).
Regarding claim 6, Schuelke teaches all of the limitations of claim 1 as noted above.
Schuelke further teaches assembling the training dataset includes processing the motion-simulated k-space data to extract features indicative of motion artifacts (Pg. 39, ln. 3-9; As for the method described in (3), this motion estimation can be performed once from the undersampled image and the synthesized contrast, or at every iteration)
Schuelke does not explicitly teach storing the extracted featured in the training dataset.
Mysore Siddu, however, teaches in a similar field of endeavor a method for training a neural network (Paragraph [0091]; Embodiments may employ traditional machine-learning, deep learning and/or image processing techniques to build motion classification models) to detect motion artifacts (Paragraph [0053]; concept for detecting the presence or absence of subject motion during a medical scan) comprising storing the extracted featured (Paragraph [0070]; system 100 also comprises a feature extraction component 120 adapted to extract an MRI feature of the MRI slice image) in the training dataset (Paragraph [0146]; detected subject motion can be stored by the data processing system (for example, in a database) and provided to other components of the system).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the method of Schuelke to have further included storing the extracted featured in the training dataset as taught by Mysore Siddu because it would have allowed cross-validation of learning schemes and further be used to generate classification models (Mysore Siddu, Paragraph [0092]). Furthermore, the extracted features can be used to determine or set thresholds for learned motion types, thereby improving the motion classification model (Paragraph [0144]-[0146]).
Claims 7-9 are rejected under 35 U.S.C. 103 as being unpatentable over Schuelke in view of Mysore Siddu as applied to claim 6 above, and further in view of Song (US 20090316971).
Regarding claim 7, together Schuelke and Mysore Siddu teach all of the limitations of claim 6 as noted above.
Together Schuelke and Mysore Siddu do not explicitly teach processing the motion-simulated k-space data to extract features indicative of motion artifacts comprises computing a cross-correlation between adjacent phase-encoding lines in the motion-simulated k-space data.
Song, however, teaches in a similar field of endeavor a method comprising extracting features indicative of motion artifacts (Paragraph [0031]-[0032]; As illustrated in FIG. 1 and FIG. 2, in EXTRACT, motion estimation; Thus, the size of the initial base is indicative of the accuracy of the estimate of the amount of motion.) comprises computing a cross-correlation (Paragraph [0041]; Here Ꚛ denotes cross correlation, Sref and S are k-space reference and acquired segments) between adjacent phase-encoding lines (Paragraph [0041]; As shown in the Appendix, for a segment size of one phase-encoding line, C(x,y) equals the correlation function of a projection along the readout (x) axis, multiplied by a phase term determined by the shift along y) in the motion-simulated k-space data (Paragraph [0031]-[0032]; a k-space extrapolation is performed along both positive and negative phase-encoding (y) directions to estimate a "motion-free" reference data in the two adjacent regions; method of k-space extrapolation provides an accurate estimation of subsequent motion).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the processing the motion-simulated k-space data of Schuelke in view of Mysore Siddu to include extracting features indicative of motion artifacts comprises computing a cross-correlation between adjacent phase-encoding lines in the motion-simulated k-space data as taught by Song because it would have allowed efficiently and quickly recovering estimates of motion information by performing a single step, and further does not rely on a range of trial transitions, thereby increasing the speed of motion artifact extraction (Song, Paragraph [0028]).
Regarding claim 8, together Schuelke, Mysore Siddu, and Song teach all of the limitations of claim 7 as noted above.
Schuelke discloses the invention as claimed and discussed above, but fails to explicitly disclose the extracted features are labeled with different severities of motion artifact based on a magnitude of the cross-correlation.
Song, however, further teaches extracted features are labeled with different severities of motion artifact based on a magnitude of the cross-correlation (Paragraph [0041]; Translational motion along the x-direction will shift the correlation maximum by the same amount, while along the y-axis there will only be a phase shift. Therefore, a maximum search operation on the real component of C(x,y) is used to recover both x and y motion.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have further modified the method of Schuelke in view of Mysore Siddu and Song such that the extracted features are labeled with different severities of motion artifact based on a magnitude of the cross-correlation because it would have allowed the network to learn the exact amount of motion and thus determine the amount of compensation needed for production of motion free images (Song, Paragraphs [0054]-[0055]) and further reduces the need to collect extra navigation data (Paragraph [0056]).
Regarding claim 9, together Schuelke, Mysore Siddu, and Song teach all of the limitations of claim 8 as noted above.
Schuelke discloses the invention as claimed and discussed above, but fails to explicitly disclose the extracted features are labeled with different severities of motion artifact based on the magnitude of the cross-correlation in a central region of the k-space.
Song, however, further teaches the extracted features are labeled with different severities of motion artifact based on the magnitude of the cross-correlation in a central region of the k-space (Paragraph [0031]; EXTRACT, motion estimation and correction are performed in a progressive manner, starting from the central k-space region and growing outward from immediate previously selected regions; paragraph [0032]; initial motionless base was chosen to consist of only 3 central ky lines).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have further modified the method of Schuelke in view of Mysore Siddu and Song such that the extracted features are labeled with different severities of motion artifact based on the magnitude of the cross-correlation in a central region of the k-space because the extrapolation technique results in smaller error in the detected motion in the central k-space locations and thereby improve image recovery of details with lower spatial frequency (Song, Paragraph [0053]).
Claims 15 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Schuelke (WO 2021228515 A1) in view of Amthor (US 20230288514).
Regarding claim 15, Schuelke teaches a method for detecting motion artifacts in k-space data (pg. 13, ln. 31-pg. 14, ln. 8; The image generating neural network is configured for outputting the synthetic magnetic resonance image data in response to receiving the reference magnetic resonance image data as input; Pg. 43, ln. 14-15; The artifact level in the resulting fused image is estimated using a dedicated motion artifact level estimator) acquired with a magnetic resonance imaging (MRI) system (Pg. 31, ln. 20-33; magnetic resonance imaging system 302, Fig. 3), the method comprising:
(a) acquiring k-space data from a subject using the MRI system (Pg. 32, ln. 32-Pg. 33, ln. 10; configured for acquiring reference k-space data 334… reference k-space data 334 and the measured k-space data 124 could be acquired at different times for the same subject 318);
(b) accessing a machine learning model with a computer system (Pg. 30, ln. 10-19; The memory 110 is further shown as containing an image generating neural network 122, Fig. 1), wherein the machine learning model has been trained (Pg. 37, ln. 16-20; Train a contrast-to-contrast network; This network can be trained using a training dataset of pairs of the same image with contrast A and B) on training data (Pg. 43, ln. 13-22; Generation of the associated training dataset was realized based on motion-free volunteer T2w images as well as an artifact simulation pipeline; Pg. 42, ln. 22-Pg. 43, ln. 5; Creation of a suitable dataset can be realized in multiple ways… the first motion-free scan is then converted to the target contrast, i.e. the contrast of a second scan that is corrupted by motion artifacts) to detect motion artifacts in k-space data (Pg. 42, Ln. 1; network that is trained to estimate the motion artifact level in an image);
(c) inputting the k-space data to the machine learning model, generating motion artifact classification data as an output ((Pg. 42, Ln. 1; network that is trained to estimate the motion artifact level in an image), wherein the motion artifact classification data indicate a presence and severity of motion artifacts in the k-space data (Pg. 43, ln. 13-22; If the estimated artifact level in the fused image is considerably lower than the estimated artifact level in the original image, the profile(s) are considered to be corrupted by motion; differentiating between having motion artifacts and not having motion artifacts is considered to read on the claimed limitation of different classifications of motion artifact severity as understood in its broadest reasonable interpretation); and
(d) analyzing the motion artifact classification data with the computer system to control operation of the MRI system (Pg. 21, ln. 27-Pg. 22, ln. 4; calculate a comparison metric between the synthetic k-space data and each of the groups of k-space data and perform a predetermined action if the comparison metric is outside of a predetermined value range… wherein the predetermined action is any one of the following: a reacquisition of at least a portion of the groups of k-space data, a halting of the acquisition of the measured k-space data, and combinations thereof; Pg. 41, ln. 30-Pg. 42, ln. 2; automate this step using a dedicated metric, e.g. based on a neural network trained to estimate the motion artifact level in an image).
Schuelke does not explicitly teach accessing a machine learning model with a computer system while the subject is still in the MRI system;
inputting the k-space data to the machine learning model while the subject is still in the MRI system; and
analyzing the motion artifact classification data with the computer system to control operation of the MRI system while the subject is still in the MRI system, such that control signals are sent to the MRI system to acquire additional data from the subject based on the motion artifact classification data output by the trained machine learning model.
Amthor, however, teaches a method for detecting motion artifacts in k-space data (Paragraph [0021]; In another embodiment, a preview image or composite image of artificial contrast is generated by summing up all complex values of the fingerprint signals… Motion artifacts will still be present in this image can be assessed automatically) acquired with a magnetic resonance imaging (MRI) system (Paragraph [0012]; The magnetic resonance imaging system is configured to arrange for reconstruction of the set of magnetic resonance images), the method comprising:
(a) acquiring k-space data from a subject using the MRI system (Paragraph [0076]; The method starts with step 400. In step 400 the MRF k-space data 332 is acquired by controlling the magnetic resonance imaging system 302);
(b) accessing a machine learning model with a computer system (Paragraph [0077]; proceeds to steps 200 and 202 as was performed in the method illustrated in FIG. 2., reconstruct MRF data and receive MRF data; Paragraph [0021]; The MRF scoring module is further configured to provide the MRF quality score by identifying motion artifact regions in the composite image. This task can be performed using AI-based detection) while the subject is still in the MRI system (Paragraph [0022]; example may be done rapidly enough that it can be performed immediately after an examination by the subject and enable reacquisition of the MRF data if necessary.);
(c) inputting the k-space data to the machine learning model while the subject is still in the MRI system (Paragraph [0077]; proceeds to steps 200 and 202 as was performed in the method illustrated in FIG. 2.), generating motion artifact classification data as an output (Paragraph [0092]-[0093]; Motion artifacts that make diagnostically important patterns less visible… use spatial frequency analysis or deep learning approaches to identify prevalence of specific motion artifacts); and
(d) analyzing the motion artifact classification data with the computer system to control operation of the MRI system while the subject is still in the MRI system (Paragraph [0077]; Next, the method proceeds to step 204, which is a decision box with the question “Is the MRF quality score 126 within the predetermined range 128; if the answer was no, then the method proceeds to step 208), such that control signals are sent to the MRI system to acquire additional data from the subject based on the motion artifact classification data output by the trained machine learning model (Paragraph [0077]; if the answer was no, then the method proceeds to step 208… In step 406 the method returns back to step 400. Essentially the signal causes the magnetic resonance imaging system to reacquire the MRF k-space data 332 and repeat the method.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have modified the method of Schuelke to have included accessing a machine learning model with a computer system while the subject is still in the MRI system; inputting the k-space data to the machine learning model while the subject is still in the MRI system; and analyzing the motion artifact classification data with the computer system to control operation of the MRI system while the subject is still in the MRI system, such that control signals are sent to the MRI system to acquire additional data from the subject based on the motion artifact classification data output by the trained machine learning mode as taught by Amthor because it would have allowed the magnetic resonance images to be evaluated more quickly for motion artifacts and thereby help reduce otherwise wasted magnetic resonance imaging system time and further enable the reacquisition of the MRF data before a subject leaves the examination room (Paragraphs [0008] and [0015]).
Regarding claim 18, together Schuelke and Amthor teach all of the limitations of claim 15 as noted above.
Schuelke further teaches the machine learning model is a