19DETAILED 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 .
Response to Amendment
The affidavit under 37 CFR 1.130(a) filed on 5/8/2026 is sufficient to overcome the rejection of claims 1-20 based on 103. The affidavit provides evidence that the additional individuals on the applied secondary reference worked under the inventors and other co-inventors direction and/or control of the invention that is disclosed in the application invention. This is considered evidence of reliance on the exception provision of 35 U.S.C. 102(b)(1).
Response to Arguments
Applicant’s arguments, see page 6, filed 5/8/2026, with respect to 112 (a) rejection have been fully considered and are persuasive. The 112 (a) rejection of the claims 4 and 19 has been withdrawn.
Applicant’s arguments, see pages 7-10, filed 10/30/2025, with respect to the rejection(s) of claim(s) 1-18 under 103 have been fully considered and are persuasive. Therefore, the rejection has been withdrawn. However, upon further consideration, a new ground(s) of rejection is made in view of Burrus.
Regarding the arguments mentioned in the specification, the remarks state that the references do not perform the features of “wherein the CDTD indicates a proportion of water molecules in each voxel in the region-of-interest whose diffusion corresponds to each of a plurality of diffusion tensors”. In particular, the arguments state that the primary reference does not teach a distribution that indicates a proportion of water molecules in each voxel whose diffusion corresponds to each of a plurality of diffusion tensors. The reference of Burrus is combined to cure this deficiency.
The reference of Burrus discloses a tensor field that comprises a plurality of tensors. In ¶ [95], a plurality of partial tensors can correspond to combined directions of brain fibers that are a micron in size smaller than the size of DTMRI voxels that are mm3 in size. It further details that the substantially anisotropic tensors combined can cause a substantially isotropic tensor to appear. However, the scanning of these combined areas will cause partial volume effects. The system further discusses in ¶ [97] the difficulty of determining fiber paths when considering crossing fibers. A tensor encodes one diffusion direction at a fiber, and at crossing fibers, multiple tensors are present at crossing points. The micron sized fibers at crossing points within the voxel sized areas can generate isotropic tensors and lead to ambiguities. The invention solution to this issue is to reduce each tensor to a scalar value and utilize scalar-valued volume visualization. Therefore, with this solution associated with the role of DT-MRI to determine the quantity of water molecules in a scanned area (see ¶ [62]-[67]) and having a volume of tensors within a voxel to describe water molecule spatial behavior (see ¶ [52]), this performs the feature of contended claim feature above.
Thus, based on the above, the features of the claims are disclosed below.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claim(s) 1-3 is/are rejected under 35 U.S.C. 103 as being unpatentable over Koay (US Pub 2009/0118608) in view of Burrus (US Pub 2006/0229856).
Re claim 1: Koay discloses a method for imaging diffusion using magnetic resonance imaging (MRI), the method comprising:
(a) accessing, with a computer system, diffusion-weighted imaging data acquired from a region-of-interest in a subject using an MRI system (e.g. a portion of interest of a subject is captured and a diffusion weighted image is acquired represented from the acquired MRI image, which is taught in ¶ [30]-[32].);
[0030] Magnetic resonance is generated and spatially encoded in at least a portion of a region of interest of the imaging subject 106. By applying selected magnetic field gradients via the gradient coils 116, a selected k-space trajectory is traversed, such as a Cartesian trajectory, a plurality of radial trajectories, or a spiral trajectory. Alternatively, imaging data may be acquired as projections along selected magnetic field gradient directions. During imaging data acquisition, a radio frequency receiver or receivers 146 coupled to the head coil 124 may acquire magnetic resonance samples that are stored in a magnetic resonance data memory 150. Of course, it is contemplated that the magnetic resonance signals may be excited and received by a single coil array such as the whole body coil 122 or the local coil 124.
[0031] The MRI sequence may include a complex series of magnetic field gradient pulses and/or sweeps generated by the gradient coils 116 which along with selected RF pulses generated by the coils 122, 124 result on magnetic resonance echoes. For diffusion tensor magnetic resonance imaging (DT-MRI), data is acquired without diffusion weighting, and with diffusion weighting in N directions. Six diffusion directions may provide sufficient information to construct the diffusion tensor at each voxel. A reconstruction engine 152 may reconstruct the imaging data into an image representation, e.g., diffusion-weighted image representations. The reconstructed image generated by the reconstruction engine 152 may be stored in an image memory 154.
[0032] A diffusion tensor engine 160 may calculate a plurality of diffusion coefficients values on a per voxel basis, as known in the art, to construct a diffusion tensor map 162. For each voxel, a tensor, e.g., a symmetric positive definite 3.times.3 matrix, that describes a three dimensional shape of diffusion may be calculated. The diffusion tensor at each voxel may be processed to obtain eigenvectors and eigenvalues.
(b) generating a comprehensive diffusion tensor distribution (CDTD) from the diffusion-weighted imaging data using the computer system, wherein the CDTD indicates a proportion in each voxel in the region-of-interest whose diffusion corresponds to each of a plurality of diffusion tensors (e.g. a diffusion tensor engine calculates diffusion coefficient values on a per voxel basis and generates a plurality of tensors associated with the 3D shape of the diffusion, which is taught in ¶ [32] and [33]. The three-dimensional shape of the diffusion for each voxel can be calculated. The diffusion tensor map expresses a proportion of diffused water. The map shows a portion or portions of diffusion within an overall area, which is taught in ¶ [34], [35] and [150]. Since the voxel shows a part of an MRI image that conveys diffusion, or movement of water, within an area, this invention performs the above limitation.); and
[0032] A diffusion tensor engine 160 may calculate a plurality of diffusion coefficients values on a per voxel basis, as known in the art, to construct a diffusion tensor map 162. For each voxel, a tensor, e.g., a symmetric positive definite 3.times.3 matrix, that describes a three dimensional shape of diffusion may be calculated. The diffusion tensor at each voxel may be processed to obtain eigenvectors and eigenvalues.
[0033] As described in a greater detail below, a covariance matrix determining engine 164 may construct a covariance matrix of, for example, the major eigenvector at each voxel. It is contemplated that covariance matrices of the medium and minor eigenvectors may be constructed as well. A cone of uncertainty (COU) determining engine 170 may construct or determine a cone of uncertainty (COU) based on the covariance matrix. A first normalized measure engine 172 may determine a first normalized measure of the elliptical COU. A second normalized measure engine 174 may determine a second normalized measure or a normalized circumference of the elliptical COU.
(c) outputting the CDTD with the computer system (e.g. the system discloses outputting the rendered map data using the tensors output associated with voxels, which is taught in ¶ [34]-[36].).
[0034] A rendering engine 176 may format the constructed cones of uncertainty, normalized areal maps, normalized circumference maps, or any other appropriate image representations to be displayed on a user interface 180, stored in non-volatile memory, transmitted over a local intranet or the Internet, viewed, stored, manipulated, or so forth.
[0035] Major eigenvector maps represent the fiber orientation and may be visualized as vector-field or color coded maps giving a cartography of the tracts' position and direction. The brightness may be weighted by a fractional anisotropy which is a scalar measure of a degree of anisotropy for a given voxel. To visualize the fiber direction map in the context of a conventional structural image, the fiber map may be registered to and superimposed on a structural image, for example, on a high resolution MR image, as known in the art. The fiber orientation information provided by the diffusion imaging maps may be used to differentiate among nerve fiber pathways to determine abnormalities. Further, the fiber orientation information provided by the diffusion imaging maps may be used to delineate anatomical structures based on a priori distinct fiber architecture of each anatomical structure.
[0036] The user interface 180 may also enable a radiologist, technician, or other operator of the magnetic resonance imaging scanner 100 to communicate with the magnetic resonance imaging controller 140 to select, modify, and execute magnetic resonance imaging sequences.
However, Koay fails to specifically teach the features of wherein the CDTD indicates a proportion of water molecules in each voxel in the region-of-interest whose diffusion corresponds to each of a plurality of diffusion tensors.
However, this is well known in the art as evidenced by Burrus. Similar to the primary reference, Burrus discloses diffusion tensors (same field of endeavor or reasonably pertinent to the problem).
Burrus discloses wherein the CDTD indicates a proportion of water molecules in each voxel in the region-of-interest whose diffusion corresponds to each of a plurality of diffusion tensors (e.g. the invention discloses diffusion tensors that provide information about water diffusion that describe the behavior of water molecules in each voxel. The diffusion tensor images can be used to corresponds to parts of the brain, which is taught in ¶ [37]-[40] and [52]-[56]. As seen in figure 6, the system can capture multiple tensors in order to determine the diffusion within the captured information, which is taught in ¶ [95] and [100]. The use of the DT-MRI is to gather information to determine the quantity of water molecules, which is taught in ¶ [63]-[67].).
[0037] Diffusion tensor images can provide information about water diffusion. The acquisition process can result in a volume of tensors where each voxel describes the local spatial behavior of water molecules. Diffusion properties can be good clues about underlying geometric structures, such as: [0038] fibers connecting different regions of the brain, which can induce anisotropic water diffusion, like in a tube; and/or [0039] gray matter, without particular structure, which can induce isotropic diffusion; etc.
[0040] Thus, by using DT-MRI data, certain exemplary embodiments can infer some information about fibers and connectivity. Certain exemplary embodiments can find accurate and efficient techniques to visualize diffusion tensor volumes.
[0052] Emphasis can be placed on the third item. Diffusion tensor images can provide information about water diffusion. An acquisition process results in a volume of tensors where each voxel can describe a local spatial behavior of water molecules.
[0053] Diffusion Properties Can be Clues About Underlying Geometric Structures, Such as: [0054] fibers connecting different regions of the brain induce highly anisotropic water diffusion, like in a tube; and/or [0055] gray matter, without particular structure, induces isotropic diffusion; etc.
[0056] Thus, by using DT-MRI data, it might be possible to infer some information about fibers and connectivity. Certain exemplary embodiments can find accurate and efficient techniques to visualize diffusion tensor volumes. Possible approaches can comprise: [0057] global visualization of the whole diffusion tensor volume; and/or [0058] individual fiber localization, using tracking algorithms; etc.
[0063] DT-MRI can be based on diffusion Magnetic Resonance Imaging (MRI), which can be based on MRI. Using nuclear spinning properties of hydrogen atoms, MRI scanners can send a radio frequency pulse to the desired area, and non-intrusively gather several information, such as: [0064] quantity of water molecules; [0065] chemical surrounding of the water (T1 relaxation); and/or [0066] chemical surrounding of each individual hydrogen atoms (T2 relaxation), which gives a different contrast than T1; etc.
[0067] By applying two symmetric additional magnetic gradients in a particular direction, Diffusion MRI can analyze the diffusion coefficient of water molecules in a gradient's direction. This diffusion coefficient can be related to a quantity of water, which moved into the gradient's direction during the acquisition time.
[0095] FIG. 6 is a diagram of exemplary diffusion tensors. Magnetic Resonance scanners can have a rather low resolution; the size of DTMRI voxels can be 1.7.times.1.7.times.3 mm.sup.3. Brain fibers can be much thinner, in the order of a micron. This can lead to partial volume effects: fibers with different directions may be regrouped within a given tensor. This is depicted in FIG. 6. The resulting tensor can be a sum of partial tensors corresponding to combined directions. A combination of two substantially anisotropic tensors can give a substantially isotropic tensor, this means that even in substantially anisotropic regions comprising fibers, substantially isotropic tensors may appear. Increasing scanner resolution can help to a certain extent, but the size order of fibers can be too small to avoid partial volume effects. In the middle row of FIG. 6, the diffusion tensors can be the average of two different directions.
[0096] DT-MRI data can be noisy. Noise can be accumulated during image acquisition and the subject himself might add thermal noise. It can be shown that MR noise follows a Rician distribution. Any of several methods can be utilized to remove this Rician noise. However, in most cases, MR noise can be modeled by a Gaussian noise, which can be a good approximation for certain signal to noise ratios.
[0097] FIG. 7 is a diagram of exemplary fibers. Two considerations can increase a difficulty of fiber path determination: crossing fibers and branching fibers. Some fibers might cross other fibers. A tensor might only encode one diffusion direction. So resulting tensors in crossing points can be a combination of two underlying fibers directions. This can be problematic. FIG. 7 shows a tensor field corresponding to crossing fibers. Robust algorithms might be utilized to cope with ambiguities in crossing points. FIG. 7 shows crossing fibers, which can generate isotropic tensors and lead to ambiguities.
[0100] In certain exemplary embodiments, no assumption about the underlying structures need be made, certain exemplary techniques can be applied to any DT-MRI dataset. A potential approach to represent tensor fields can comprise reducing each tensor to a scalar value, and then utilizing scalar-valued volume visualization. In DT-MRI, information can comprise a degree of anisotropy. In certain exemplary methods, a degree of anisotropy can depend on a chosen gradient direction. Fractional Anisotropy Factional Anisotropy (FA), can be defined by: FA = 3 2 * ( .lamda. 1 - .lamda. _ ) 2 + ( .lamda. 2 - .lamda. _ ) 2 + ( .lamda. 3 - .lamda. _ ) 2 .lamda. 1 2 + .lamda. 2 2 + .lamda. 3 2 ##EQU4## [0101] where .lamda..sub.1, .lamda..sub.2, .lamda..sub.3 are eigenvalues and {overscore (.lamda.)} is the mean eigenvalue: .lamda. _ = .lamda. 1 + .lamda. 2 + .lamda. 3 3 . ##EQU5##
[0112] Instead of considering a whole tensor, certain exemplary embodiments can reduce the tensor to one vector, comprising a principal diffusion direction (PDD). By reducing each tensor to a main eigenvector, tensor field visualization can be simplified into a vector field or flow visualization problem. Line Integral Convolution (LIC) can be a flow visualization algorithm. Texture-based, LIC can accurately render high curvature details and generate high resolution images. In addition, LIC can be robust to noisy data.
Therefore, in view of Burrus, it would have been obvious to one of ordinary skill before the effective filing date of the claimed invention was made to have the feature of wherein the CDTD indicates a proportion of water molecules in each voxel in the region-of-interest whose diffusion corresponds to each of a plurality of diffusion tensors, incorporated in the device of Koay, in order to use tensors to provide information about water diffusion, which can improve understanding of brain internals (as stated in Burrus ¶ [32]-[37]).
Re claim 2: Koay discloses the method of claim 1, wherein generating the CDTD comprises defining a maximum diffusion tensor value, defining a minimum diffusion tensor value, and constraining the plurality of diffusion tensors by the maximum diffusion tensor value and the minimum diffusion tensor value (e.g. the diffusion tensor values contain a largest value and a smallest value. The tensors are between these different values, which is taught in ¶ [37].).
[0037] With continuing reference to FIG. 1A and further reference to FIGS. 1B and 1C, an eigenvector/eigenvalue ordering engine 182 may order the diffusion tensor eigenvectors and eigenvalues for each voxel to obtain ordered eigenvectors and eigenvalues. For example, the eigenvector/eigenvalue ordering engine 182 may order eigenvalues from largest to smallest eigenvalue .lamda.1, .lamda.2, .lamda.3. The eigenvector which corresponds to the largest eigenvalue .lamda.1 is called the major eigenvector, and aligns with the spatial direction having the highest diffusion coefficient. The medium and smallest eigenvalues .lamda.2, .lamda.3 correspond to the medium and minor eigenvectors. The medium and minor eigenvectors are orthogonal to the major eigenvector and align with spatial directions having lower diffusion coefficients. The relative values of the eigenvalues .lamda.1, .lamda.2, .lamda.3 are indicative of the spatial orientation and magnitude of the diffusion tensor anisotropy.
[0038] As shown in FIG. 1C, the eigenvectors and eigenvalues are geometrically representable by an ellipsoid 183 whose long axis aligns with the major eigenvector e1, i.e. with the direction of the highest apparent diffusion coefficient. The deviation of the ellipsoid 183 from a perfect sphere is representative of the anisotropy of the diffusion tensor. An anisotropic diffusion coefficient tensor may reflect the influence of neural fiber bundles 184 which tend to inhibit diffusion in directions partially or totally orthogonal to the fibers 184, e.g. the directions of eigenvectors e2, e3. In contrast, diffusion parallel to the fibers 184, i.e. channeled along the direction of the major eigenvector e1, is enhanced and larger than along the e2, e3 directions.
Re claim 3: Koay discloses the method of claim 1, wherein each of the plurality of diffusion tensors is described by eigenvalues, λ1, λ2, and λ3 and wherein generating the CDTD comprises constraining the plurality of diffusion tensors by λ1
PNG
media_image1.png
16
51
media_image1.png
Greyscale
PNG
media_image2.png
17
20
media_image2.png
Greyscale
. (e.g. the diffusion tensor values contain a largest value and a smallest value. The tensors are between these different values, which is taught in ¶ [37] above. The diffusion tensor values are described as eigenvalues.).
Claim(s) 4 is/are rejected under 35 U.S.C. 103 as being unpatentable over Koay, as modified by Burrus, as applied to claim 1 above, and further in view of Frank (US Pub 2020/0386839) and Arsigny (US Pub 2008/0170802).
Re claim 4: However, Koay fails to specifically teach the features of the method of claim 1, wherein generating the CDTD comprises selecting the plurality of diffusion tensors by logarithmically discretizing a range of candidate eigenvalues for each of the plurality of diffusion tensors.
However, this is well known in the art as evidenced by Frank. Similar to the primary reference, Frank discloses estimating local diffusion anisotropy (same field of endeavor or reasonably pertinent to the problem).
Frank discloses wherein generating the CDTD comprises selecting the plurality of diffusion tensors by discretizing a range of candidate eigenvalues for each of the plurality of diffusion tensors (e.g. the system discloses discretizing a range of eigenvalues for a tensor. This is performed for the tensor selected for processing. This is disclosed in ¶ [52], [64], [67] and [68].).
[0052] The radial and angular variances σ.sub.R.sub.1.sub.,R.sub.2 and σ.sub.Ω.sub.1.sub.,Ω.sub.2 characterize diffusion properties at different radial and angular scales, respectively, and as such contain a great deal of information. However, they are still multidimensional tensors and from a practical perspective it is useful to derive average quantities of lower dimensionality, particularly scalar quantities that can be easily overlaid on images to more easily discern spatial variations in diffusion properties that correspond to particular brain regions and tissue types.
[0064] An illustrative example of DiTSI is provided by simulating the signal from a single voxel with structures composed of idealized WM and GM geometries. The voxel diffusion characteristics were constructed in a similar fashion as disclosed by Özarslan, et al., J. Chem Phys. 2009; 130:104702 (incorporated herein by reference) by assuming that the voxel is discretized into p equal sections in each direction and thus p.sup.3 cubic subvolumes (or “pores”), each of which contained a simple ellipsoidal diffusion profile characterized by a real symmetric 3×3 diffusion tensor characterized by the three principal eigenvectors and their associated eigenvalues {λ.sub.1, λ.sub.2, λ.sub.3}. The simulation input allowed arbitrary manipulation of the relative eigenstructure of these ellipsoids, facilitating the generation of various “tissue” models. For illustrative purposes, mitigation of angular sampling discretization provides more informative examples and so a large number of unique diffusion angles (100) generated uniformly on a sphere were used, although the simulation program can also use any dPFG strategy such as those shown in FIG. 11. (Note that the sampling in the octahedron has orthogonal vectors while the cube does not.) For the dPFG simulation two b-values were used and the complete sampling is generated as the Cartesian product of the gradient pattern, resulting in 40 k samples. For the simulations shown below, p=6 giving a total of 6.sup.3=216 pores in the “voxel”. For the acquisition parameters, we used a diffusion time Δ=50 ms, width of diffusion gradients δ=20 ms, and two shells with b=(2401, 4803)s/mm.sup.2.
[0067] The intensity pattern of σ.sub.R.sub.1.sub.,R.sub.2 reveals an interesting and potentially useful new aspect of the DiTSI analysis. Note that the fall-off with larger radius is not monotonic but has peaks. This is most likely a result of both a limited range of eigenvalues used for the dPFG signal generation and a fixed sampling/discretization size used for the data acquisition and analysis. These results nonetheless demonstrate the ability of DiTSI to distinguish between different kinds of microstructure organization and possible development of more detailed metrics of tissue microstructure. Incorporation of more detailed physical models will facilitate understanding of the significant and application of this structure. With the recognition that this is not “diffusion diffraction” in the traditional sense, these intensity profile variations are nevertheless related to the pore geometry and so will be referred to as “diffraction-like” intensity patterns.
[0068] FIG. 2 illustrates the GM tissue model simulation results. In the first model (GM.sub.1), the local diffusion profiles are identical to the WM model but with random orientations. In the second model (GM.sub.2), all three eigenvalues are the same in each pore (i.e., no anisotropy), but different between pores. The third model (GM.sub.3) combines the first two, with random orientations and variations in the relative magnitude of the eigenvalues in each pore, but still ellipsoidal. The Φ estimated from an sPFG acquisition (Row 2) is approximately spherical, reflecting the fact that sPFG is sensitive only to macroscopic anisotropy, which these voxels do not possess, but not to their existing microscopic anisotropy. The angular variance σ.sub.Ω.sub.1 (Equation (15)) in Row 3 demonstrates the ability of DiTSI to detect the microscopic angular variations in both GM.sub.1 and GM.sub.3 with greater intensities representing directions of higher average microscopic anisotropy. In the case with no angular variation (GM.sub.2), the angular variance is zero. In Row 4 are shown the DiTSI results of the full radial variance σ.sub.R.sub.1.sub.,R.sub.2 (Equation 6). In both models that have ellipsoid diffusion profiles (GM.sub.1 and GM.sub.3), and hence a continuous (though bounded) range of radial dimensions, the results are similar to those FIG. 1, with a pronounced peak at the origin R.sub.1=R.sub.2 and falloff consistent with the diffraction-like intensity profile variations related to the pore geometry. In GM.sub.2, where only the radius is changed, only certain components consistent with the relatively few discretized radial variations are apparent. These results demonstrate the ability of DiTSI to detect both angular and radial variations in microscopic anisotropy.
Therefore, in view of Frank, it would have been obvious to one of ordinary skill before the effective filing date of the claimed invention was made to have the feature of wherein generating the CDTD comprises selecting the plurality of diffusion tensors by discretizing a range of candidate eigenvalues for each of the plurality of diffusion tensors, incorporated in the device of Koay, in order to discretize eigenvalues associated with each tensor, which aids in the determination of quantitative diffusion anisotropy and connectivity measures (as stated in Frank ¶ [12]).
However, the combination above fails to specifically teach the feature of logarithmically discretizing.
However, this is well known in the art as evidenced by Arsigny. Similar to the primary reference, Arsigny discloses diffusion tensors (same field of endeavor or reasonably pertinent to the problem).
Arsigny discloses logarithmically discretizing (e.g. the system discloses eigenvalues that are in logarithmic form, which is taught in ¶ [50]. With the previous reference discloses the discretizing of the eigenvalues combined with these eigenvalues in log form, the feature of the claim is disclosed.).
[0050] Each tensor S is associated with a unique logarithm L such as S=exp(L), where exp is the exponential of matrices. This logarithm is simply any symmetrical matrix, and conversely any symmetrical matrix corresponds to a unique tensor given by the exponential of that matrix. In an orthonormal base in which S is diagonal, L is obtained simply by transforming the eigenvalues of S into their standard logarithm, e.g. Naperian logarithm.
[0051] The theoretical framework in an application to the Naperian logarithm "log" and to its exponential reciprocal function "exp" is described below. In this application, the metric will be called "Log-Euclidian". However, this framework is not limited to the functions log and exp. It also applies to functions with a logarithmic base, and in particular to any base a logarithms and their reciprocals the any base a exponentials, as well as to other types of functions, such as for example the function "-log" or functions translated from any base a logarithms "C+/-log.sub.a", where C is a constant.
Therefore, in view of Arsigny, it would have been obvious to one of ordinary skill before the effective filing date of the claimed invention was made to have the feature of logarithmically discretizing, incorporated in the device of Koay, as modified by Frank, in order to have log eigenvalues utilized for the system, which can aid reducing noise between tensor images while retaining anisotropy data (as stated in Arsigny ¶ [63]).
Claim(s) 19 is/are rejected under 35 U.S.C. 103 as being unpatentable over Koay, as modified by Burrus, as applied to claim 1 above, and further in view of Frank (US Pub 2020/0386839).
Re claim 19: (New) However, Koay fails to specifically teach the features of the method of claim 1, wherein generating the CDTD comprises selecting the plurality of diffusion tensors by uniformly discretizing a range of candidate eigenvalues for each of the plurality of diffusion tensors.
However, this is well known in the art as evidenced by Frank. Similar to the primary reference, Frank discloses estimating local diffusion anisotropy (same field of endeavor or reasonably pertinent to the problem).
Frank discloses wherein generating the CDTD comprises selecting the plurality of diffusion tensors by uniformly discretizing a range of candidate eigenvalues for each of the plurality of diffusion tensors (e.g. the system discloses discretizing a range of eigenvalues for a tensor. This is performed for the tensor selected for processing. This is disclosed in ¶ [52], [64], [67] and [68] above.).
Therefore, in view of Frank, it would have been obvious to one of ordinary skill before the effective filing date of the claimed invention was made to have the feature of wherein generating the CDTD comprises selecting the plurality of diffusion tensors by uniformly discretizing a range of candidate eigenvalues for each of the plurality of diffusion tensors, incorporated in the device of Koay, in order to discretize eigenvalues associated with each tensor, which aids in the determination of quantitative diffusion anisotropy and connectivity measures (as stated in Frank ¶ [12]).
Claim(s) 20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Koay, as modified by Burrus, as applied to claim 1 above, and further in view of Arsigny (US Pub 2008/0170802).
Re claim 20: (New) However, Koay fails to specifically teach the features of the method of claim 1, wherein at least one of the plurality of diffusion tensors is not constrained according to its shape; and wherein the at least one of the plurality of diffusion tensors is not constrained according to its symmetry.
However, this is well known in the art as evidenced by Arsigny. Similar to the primary reference, Arsigny discloses diffusion tensors (same field of endeavor or reasonably pertinent to the problem).
Arsigny discloses wherein at least one of the plurality of diffusion tensors is not constrained according to its shape; and wherein the at least one of the plurality of diffusion tensors is not constrained according to its symmetry (e.g. the system discloses zero or negative eigenvalues that are not constrained to a particular shape or its symmetry, which is taught in ¶ [57], [58] and 126].).
[0057] Contrary to the conventional Euclidian framework for tensors, it can be seen from the equation (1) that the symmetrical matrices with negative or zero eigenvalues are at infinite distance from the tensors and in practice do not appear in the calculations.
[0058] Moreover, the Log-Euclidian distances are invariant by inversion: the principle of symmetry between a tensor and its inverse is verified. Consequently, the mean tensor Log-Euclidian is a generalisation of the geometric mean and not of the arithmetic mean given by the conventional Euclidian framework. In particular, the mean Log-Euclidian of a tensor and its inverse is identity. This is crucial in particular in order to obtain a correct interpolation of the determinants when two tensors are interpolated.
[0126] The second processing unit MT2' is in this case intended to carry out one or more processes on initial raw data representing an initial raw vector image or initial symmetrical matrices (but not necessarily defined as positive) which do not come from a first processing device D according to the invention.
[0127] The data to be processed in this case come direct from an imager (or image capture device). These are not therefore data forming a second representation R2 resulting from the processing performed by the first processing unit MT1 on a first representation R1 of an initial tensor image.
Therefore, in view of Arsigny, it would have been obvious to one of ordinary skill before the effective filing date of the claimed invention was made to have the feature of wherein at least one of the plurality of diffusion tensors is not constrained according to its shape; and wherein the at least one of the plurality of diffusion tensors is not constrained according to its symmetry, incorporated in the device of Koay, as modified by Burrus, in order to contain diffusion tensors that are not constrained to its symmetry or its shape, which may be further used to develop supplementary data to aid in improving processing initial data (as stated in Arsigny ¶ [33]).
Claim(s) 5, 6 and 11-14 is/are rejected under 35 U.S.C. 103 as being unpatentable over Koay, as modified by Burrus, as applied to claim 1 above, and further in view of Xie (US Pub 2022/0230310).
Re claim 5: However, Koay fails to specifically teach the features of the method of claim 1, wherein outputting the CDTD comprises generating a water pool image from the CDTD and outputting the water pool image with the computer system, wherein the water pool image depicts water diffusion associated with a subset of the plurality of diffusion tensors.
However, this is well known in the art as evidenced by Xie. Similar to the primary reference, Xie discloses classifying an image using a clustering algorithm (same field of endeavor or reasonably pertinent to the problem). Xie discloses wherein outputting the CDTD comprises generating a water pool image from the CDTD and outputting the water pool image with the computer system, wherein the water pool image depicts water diffusion associated with a subset of the plurality of diffusion tensors (e.g. the invention discloses displaying a representation of water diffusion using the diffusion tensor imaging concept. This technique involves using tensors to infer, in each voxel, different diffusion rates, which is taught in ¶ [61]-[64].).
[0061] After the initial focus on tissue (proton) density function and tissue relaxation properties researchers explored other methods to generate contrast exploiting other properties of water molecules. Diffusion imaging (DI) was a result of those research efforts. In DI, supplemental MR gradients are applied during the image acquisition. The motion of protons during the application of these gradients affects the signal in the image, thereby, providing information on molecular diffusion. DI may be performed using several techniques including diffusion-spectrum imaging (DSI) and diffusion-weighted imaging (DWI).
[0062] DWI is a non-invasive imaging method, with sensitivity to water diffusion within the architecture of the tissues that uses existing MRI technology in combination with specialized software and requires no additional hardware equipment, contrast agents, or chemical tracers. To measure diffusion using MRI, the supplemental MR gradients are employed to create an image that is sensitized to diffusion in a particular direction. In DWI, the intensity of each image element (voxel) reflects the best estimate of the rate of water diffusion in the particular direction. However, biological tissues are highly anisotropic, meaning that their diffusion rates are not the same in every direction. For routine DWI, the anisotropic nature of tissue is often ignored and the diffusion is reduced to a single average value, the apparent diffusion coefficient (ADC), but this is overly simplistic for many use cases. An alternative method is to model diffusion in complex materials using a diffusion tensor, a [3×3] array of numbers corresponding to diffusion rates in each combination of directions. The three diagonal elements (Dxx, Dyy, Dzz) represent diffusion coefficients measured along each of the principal (x-, y- and z-) laboratory axes. The six off-diagonal terms (Dxy, Dyz, etc) reflect the correlation of random motions between each pair of principal directions.
[0063] The introduction of the diffusion tensor model enables the indirect measurement of the degree of anisotropy and structural orientation that characterizes diffusion tensor imaging (DTI). The basic concept behind DTI is that water molecules diffuse differently along the tissues depending on its type, integrity, architecture, and presence of barriers, giving information about its orientation and quantitative anisotropy. With DTI analysis it is possible to infer, in each voxel, properties such as the molecular diffusion rate (Mean Diffusivity (MD) or Apparent Diffusion Coefficient (ADC)), the directional preference of diffusion (Fractional Anisotropy (FA)), the axial diffusivity (AD) (diffusion rate along the main axis of diffusion), and radial diffusivity (RD) (rate of diffusion in the transverse direction). DTI is usually displayed by either condensing the information contained in the tensor into one number (a scalar), or into 4 numbers (to give an R,G,B color and a brightness value, which is known as color fractional anisotropy). The diffusion tensor can also be viewed using glyphs, which are small three dimensional (3D) representations of the major eigenvector or whole tensor.
[0064] Similar to MRI and DTI, other modalities of medical imaging such as CT, X-ray, positron emission tomography (PET), photoacoustic tomography (PAT), sonography, combinations thereof such as PET-CT, PET-MR, and the like rely on various measurements, algorithms, and agents to generate image contrast and spatial resolution. For example, CT and X-ray use x-ray absorption to differentiate between air, soft tissue, and dense structures such as bone. Dense structures within the body stop x-rays and, thus, are easily imaged and visualized, whereas soft tissues vary in their ability to stop x-rays and, thus, may be faint or difficult to image and be visualized. One technique to increase image contrast in X-rays or CT scans is to utilize contrast agents that contain substances better at stopping x-rays making them more visible on an X-ray or CT image and, thus can be used to better visualize soft tissues such as blood vessels. PET uses small amounts of radioactive materials called radiotracers that can be detected and measured in a scan. The measurement differences between areas accumulating or labeled with the radiotracers versus non-accumulating or non-labeled areas is used to generate contrast to visualize structures and functions within the subject. PAT is a imaging modality based on the photoacoustic (PA) effect. A short-pulsed light source is typically used to irradiate the tissue, resulting in broadband PA waves. Following absorption of the light, an initial temperature rise induces a pressure rise, which propagates as a photoacoustic wave and is detected by an ultrasonic transducer to image optical absorption contrast. Ultrasound is a non-invasive diagnostic technique used to image inside the body. A transducer sends out a beam of sound waves into the body. The sound waves are reflected back to the transducer by boundaries between tissues in the path of the beam (e.g. the boundary between fluid and soft tissue or tissue and bone). When these echoes hit the transducer, the echoes generate electrical signals that are sent to the ultrasound scanner. Using the speed of sound and the time of each echo's return, the scanner calculates the distance from the transducer to the tissue boundary. These distances are then used to generate contrast to visualize tissues and organs.
Therefore, in view of Xie, it would have been obvious to one of ordinary skill at the time the invention was made to have the feature of wherein outputting the CDTD comprises generating a water pool image from the CDTD and outputting the water pool image with the computer system, wherein the water pool image depicts water diffusion associated with a subset of the plurality of diffusion tensors, incorporated in the device of Koay, as modified by Burrus, in order to classify tensors and outputting the classified image, which can decrease the complexity of the image data and focus the machine learning on specific tasks for more efficient use (as stated in Xie ¶ [50]).
Re claim 6: However, Koay fails to specifically teach the features of the method of claim 5, wherein the water pool image is generated by masking the CDTD to select the subset of the plurality of diffusion tensors.
However, this is well known in the art as evidenced by Xie. Similar to the primary reference, Xie discloses classifying an image using a clustering algorithm (same field of endeavor or reasonably pertinent to the problem). Xie discloses wherein the water pool image is generated by masking the CDTD to select the subset of the plurality of diffusion tensors (e.g. the invention discloses masking applied to tensors in order to specify a part of the tensors that have the mask, which is taught in ¶ [18], [19], [58], [81], [82] and [99]. The mask is used to select portions of the voxels.).
[0018] In some embodiments, the segmentation mask is determined and the determining the segmentation mask comprises: identifying a seed location of the object of interest using the sets of pixels or voxels assigned with the object class; growing the seed location by projecting the seed location towards a z-axis representing depth of the segmentation mask; and determining the segmentation mask based on the projected seed location.
[0019] In some embodiments, determining the segmentation mask further comprises performing morphological closing and filling on the segmentation mask.
[0058] As used herein, a “segmentation boundary” refers to an estimated perimeter of an object within an image. A segmentation boundary may be generated during a segmentation process where features of the image are analyzed to determine locations of the edges of the object. The segmentation boundary may further be represented by a mask such as a binary mask.
[0081] The segmenting of MRI images is split into two parts. A first part of the segmenting pertains to a first vision model constructed to perform localization (object detection) of classes within a first image (e.g., a diffusion tensor parametric map or a CT image). These classes are “semantically interpretable” and correspond to real-world categories such as the liver, the kidney, the heart, and the like. The localization is executed using EM, you only look once (YOLO) or (YOLOv2) or (YOLOv3), or similar object detection algorithms, which is initialized with a standard clustering technique (e.g., k-means clustering technique, Otsu's method, Density-Based Spatial Clustering of Applications with Noise (DBSCAN) technique, mini-batch K-means technique, or the like) heuristically. The initialization is used to provide the initial estimate of the parameters of the likelihood model for each class. Expectation maximization is an iterative process to find (local) maximum likelihood or maximum a posteriori (MAP) estimates of parameters in one or more statistical models. The EM iteration alternates between performing an expectation (E) step, which creates a function for the expectation of the log-likelihood evaluated using the current estimate for the parameters, and a maximization (M) step, which computes parameters maximizing the expected log-likelihood found on the E step. These parameter-estimates are then used to determine the distribution of the latent variables in the next E step. The result of the localization is boundary boxes or segmentation masks around each object with a probability for each class. The general location of an object of interest (e.g., the kidney) is isolated using one or more of the classes associated with the object of interest.
[0082] To alleviate the background effect of medical images, the boundary box or segmentation mask for the object of interest is projected in a slice direction (axial, coronal, and sagittal) onto a second image (e.g., a diffusion tensor parametric map or a CT image). In instances of localization being used to determine a segmentation mask, the boundaries of the projected segmentation mask are used to define a bounding box in the second image around the general location of the object of interest (e.g., the kidney) (a rectangular box drawn completely around a pixel wise mask associated with the object of interest). In some instances, the bounding box (determined via localization or defined based on boundaries of the projected segmentation mask) is enlarged by a predetermined number of pixels on all sides to ensure coverage of the object of interest. The area within the bounding box is then cropped from the second image to obtain apportion of the second image having the object of interest, and the portion of the second images is used as input into a second vision model to segment the object of interest. The second part of the of the segmenting pertains to a second vision model (a deep learning neural network) constructed with a weighted loss function (e.g., a Dice loss) to overcome the unbalanced nature between the object of interest and the background, and thus focus training on evaluating segmenting the object of interest. Moreover, the second vision model may be trained using an augmented data set such that the deep learning neural network is capable of being trained on a limited set of medical images. The trained second vision model takes as input the cropped portion of the second image and outputs the portion of the second image with an estimated segmentation boundary around the object of interest. The estimated segmentation boundary may be used to calculate a volume, surface area, axial dimensions, largest axial dimension, or other size-related metrics of the object of interest. Any one or more of these metrics may, in turn, be used alone or in conjunction with other factors to determine a diagnosis and/or a prognosis of a subject.
[0099] The localization controller 170 includes processes for localizing, using the one or more object detection models 160, an object of interest within the image 135. The localizing includes: (i) locating and classifying, using object detection models 160, objects within the first image having the first characteristic into a plurality of object classes, where the classifying assigns sets of pixels or voxels of the first image into one or more of the plurality of object classes; and (ii) determining, using the object detection models 160, a bounding box or segmentation mask for the object of interest within the first image based on sets of pixels or voxels assigned with an object class of the plurality of object classes. The object detection models 160 utilize one or more object detection algorithms in order to extract statistical features used to locate and label objects within the first image and predict a bounding box or segmentations mask for the object of interest.
Therefore, in view of Xie, it would have been obvious to one of ordinary skill at the time the invention was made to have the feature of wherein the water pool image is generated by masking the CDTD to select the subset of the plurality of diffusion tensors, incorporated in the device of Koay, as modified by Burrus, in order to classify tensors and outputting the classified image, which can decrease the complexity of the image data and focus the machine learning on specific tasks for more efficient use (as stated in Xie ¶ [50]).
Re claim 11: However, Koay fails to specifically teach the features of the method of claim 1, wherein outputting the CDTD comprises generating a classification image from the CDTD and outputting the classification image with the computer system, wherein the classification image depicts classified regions in the region-of-interest that are classified based on subsets of the plurality of diffusion tensors.
However, this is well known in the art as evidenced by Xie. Similar to the primary reference, Xie discloses classifying an image using a clustering algorithm (same field of endeavor or reasonably pertinent to the problem). Xie discloses wherein outputting the CDTD comprises generating a classification image from the CDTD and outputting the classification image with the computer system, wherein the classification image depicts classified regions in the region-of-interest that are classified based on subsets of the plurality of diffusion tensors (e.g. the invention discloses receiving a diffusion tensor parametric map to segment the image. A clustering algorithm is used to classify parts of the image as a category of parts of the body. The classification can be around an area within an image that is based on the set of tensors within a tensor map, which is taught in ¶ [81] and [82]. The information is displayed with an area on an image classified as the region of interest, which is taught in ¶ [125].).
[0081] The segmenting of MRI images is split into two parts. A first part of the segmenting pertains to a first vision model constructed to perform localization (object detection) of classes within a first image (e.g., a diffusion tensor parametric map or a CT image). These classes are “semantically interpretable” and correspond to real-world categories such as the liver, the kidney, the heart, and the like. The localization is executed using EM, you only look once (YOLO) or (YOLOv2) or (YOLOv3), or similar object detection algorithms, which is initialized with a standard clustering technique (e.g., k-means clustering technique, Otsu's method, Density-Based Spatial Clustering of Applications with Noise (DBSCAN) technique, mini-batch K-means technique, or the like) heuristically. The initialization is used to provide the initial estimate of the parameters of the likelihood model for each class. Expectation maximization is an iterative process to find (local) maximum likelihood or maximum a posteriori (MAP) estimates of parameters in one or more statistical models. The EM iteration alternates between performing an expectation (E) step, which creates a function for the expectation of the log-likelihood evaluated using the current estimate for the parameters, and a maximization (M) step, which computes parameters maximizing the expected log-likelihood found on the E step. These parameter-estimates are then used to determine the distribution of the latent variables in the next E step. The result of the localization is boundary boxes or segmentation masks around each object with a probability for each class. The general location of an object of interest (e.g., the kidney) is isolated using one or more of the classes associated with the object of interest.
[0082] To alleviate the background effect of medical images, the boundary box or segmentation mask for the object of interest is projected in a slice direction (axial, coronal, and sagittal) onto a second image (e.g., a diffusion tensor parametric map or a CT image). In instances of localization being used to determine a segmentation mask, the boundaries of the projected segmentation mask are used to define a bounding box in the second image around the general location of the object of interest (e.g., the kidney) (a rectangular box drawn completely around a pixel wise mask associated with the object of interest). In some instances, the bounding box (determined via localization or defined based on boundaries of the projected segmentation mask) is enlarged by a predetermined number of pixels on all sides to ensure coverage of the object of interest. The area within the bounding box is then cropped from the second image to obtain apportion of the second image having the object of interest, and the portion of the second images is used as input into a second vision model to segment the object of interest. The second part of the of the segmenting pertains to a second vision model (a deep learning neural network) constructed with a weighted loss function (e.g., a Dice loss) to overcome the unbalanced nature between the object of interest and the background, and thus focus training on evaluating segmenting the object of interest. Moreover, the second vision model may be trained using an augmented data set such that the deep learning neural network is capable of being trained on a limited set of medical images. The trained second vision model takes as input the cropped portion of the second image and outputs the portion of the second image with an estimated segmentation boundary around the object of interest. The estimated segmentation boundary may be used to calculate a volume, surface area, axial dimensions, largest axial dimension, or other size-related metrics of the object of interest. Any one or more of these metrics may, in turn, be used alone or in conjunction with other factors to determine a diagnosis and/or a prognosis of a subject.
[0125] At block 440 the portion of the second image with the estimated segmentation boundary around the object of interest is outputted. In some instances, the portion of the second image is provided. For example, the portion of the second image may be stored in a storage device and/or displayed on a user device.
Therefore, in view of Xie, it would have been obvious to one of ordinary skill at the time the invention was made to have the feature of wherein outputting the CDTD comprises generating a classification image from the CDTD and outputting the classification image with the computer system, wherein the classification image depicts classified regions in the region-of-interest that are classified based on subsets of the plurality of diffusion tensors, incorporated in the device of Koay, as modified by Burrus, in order to classify tensors and outputting the classified image, which can decrease the complexity of the image data and focus the machine learning on specific tasks for more efficient use (as stated in Xie ¶ [50]).
Re claim 12: However, Koay fails to specifically teach the features of the method of claim 11, wherein generating the classification image comprises inputting the CDTD to a clustering algorithm, generating the classified image as an output, wherein each of the classified regions in the region-of-interest correspond to a different cluster of voxels generated by the clustering algorithm.
However, this is well known in the art as evidenced by Xie. Similar to the primary reference, Xie discloses classifying an image using a clustering algorithm (same field of endeavor or reasonably pertinent to the problem). Xie discloses wherein generating the classification image comprises inputting the CDTD to a clustering algorithm, generating the classified image as an output, wherein each of the classified regions in the region-of-interest correspond to a different cluster of voxels generated by the clustering algorithm (e.g. the classifying of the tensors includes using a clustering algorithm that categorizes the part of the tensor or voxels. The region of interest corresponds to a voxel identifying a part of the image into a particular class, which is taught in ¶ [99] and [100].).
[0099] The localization controller 170 includes processes for localizing, using the one or more object detection models 160, an object of interest within the image 135. The localizing includes: (i) locating and classifying, using object detection models 160, objects within the first image having the first characteristic into a plurality of object classes, where the classifying assigns sets of pixels or voxels of the first image into one or more of the plurality of object classes; and (ii) determining, using the object detection models 160, a bounding box or segmentation mask for the object of interest within the first image based on sets of pixels or voxels assigned with an object class of the plurality of object classes. The object detection models 160 utilize one or more object detection algorithms in order to extract statistical features used to locate and label objects within the first image and predict a bounding box or segmentations mask for the object of interest.
[0100] In some instances, the localizing is executed using EM, YOLO, YOLOv2, YOLOv3, or similar object detection algorithms, which is initialized with a standard clustering technique (e.g., K-means or Otsu's method) heuristically. The initialization is used to provide the initial estimate of the parameters of the likelihood model for each class. For example in the instance of using EM with a K-means clustering technique, given a fixed number of k clusters, observations are assigned to the k clusters so that the means across clusters (for all variables) are as different from each other as possible. The EM clustering technique then computes posterior probabilities of cluster memberships and cluster boundaries based on one or more prior probability distributions parametrized with the initial estimate of the parameters for each cluster (class). The goal of the EM clustering technique then is to maximize the overall probability or likelihood of the data, given the (final) clusters. The results of the EM clustering technique are different from those computed by K-means clustering technique. The K-means clustering technique will assign observations (pixels or voxels such as pixel or voxel intensities) to clusters to maximize the distances between clusters. The EM clustering technique does not compute actual assignments of observations to clusters, but classification probabilities. In other words, each observation belongs to each cluster with a certain probability. Thereafter, observations may be assigned, by the localization controller 170, to clusters based on the (largest) classification probability. The result of the localization is bounding boxes or segmentation masks with a probability for each class. The general location of an object of interest (e.g., the kidney) is isolated based on sets of pixels or voxels assigned with an object class associated with the object of interest.
Therefore, in view of Xie, it would have been obvious to one of ordinary skill at the time the invention was made to have the feature of wherein generating the classification image comprises inputting the CDTD to a clustering algorithm, generating the classified image as an output, wherein each of the classified regions in the region-of-interest correspond to a different cluster of voxels generated by the clustering algorithm, incorporated in the device of Koay, as modified by Burrus, in order to classify tensors and outputting the classified image, which can decrease the complexity of the image data and focus the machine learning on specific tasks for more efficient use (as stated in Xie ¶ [50]).
Re claim 13: However, Koay fails to specifically teach the features of the method of claim 12, wherein the clustering algorithm comprises one of a hierarchical clustering algorithm, a k-means clustering algorithm, a bi-clustering algorithm, or a machine learning model-based clustering algorithm.
However, this is well known in the art as evidenced by Xie. Similar to the primary reference, Xie discloses classifying an image using a clustering algorithm (same field of endeavor or reasonably pertinent to the problem). Xie discloses wherein the clustering algorithm comprises one of a hierarchical clustering algorithm, a k-means clustering algorithm, a bi-clustering algorithm, or a machine learning model-based clustering algorithm (e.g. a k-means technique can be used in the clustering technique for classification, which is taught in ¶ [99] and [100] above.).
Therefore, in view of Xie, it would have been obvious to one of ordinary skill at the time the invention was made to have the feature of wherein the clustering algorithm comprises one of a hierarchical clustering algorithm, a k-means clustering algorithm, a bi-clustering algorithm, or a machine learning model-based clustering algorithm, incorporated in the device of Koay, as modified by Burrus, in order to classify tensors and outputting the classified image, which can decrease the complexity of the image data and focus the machine learning on specific tasks for more efficient use (as stated in Xie ¶ [50]).
Re claim 14: However, Koay fails to specifically teach the features of the method of claim 13, wherein the clustering algorithm is the machine learning model-based clustering algorithm and the machine learning model-based clustering algorithm comprises a neural network that has been trained on training data to receive CDTD data as an input and generate clusters of similar voxels as an output.
However, this is well known in the art as evidenced by Xie. Similar to the primary reference, Xie discloses classifying an image using a clustering algorithm (same field of endeavor or reasonably pertinent to the problem). Xie discloses wherein the clustering algorithm is the machine learning model-based clustering algorithm and the machine learning model-based clustering algorithm comprises a neural network that has been trained on training data to receive CDTD data as an input and generate clusters of similar voxels as an output (e.g. a CNN is used to output voxels that are classified, which is taught in ¶ [111] and [112]. The CNN is trained by input images and discussed in ¶ [87]-[89].).
[0111] The expansive path 310 is a CNN network that combines the feature and spatial information from the contracting path 305 (upsampling of the feature map from the contracting path 305). As described herein, the output of three-dimensional segmentation is not just a class label or bounding box parameters. Instead, the output (the three-dimensional segmentation mask) is a complete image (e.g., a high resolution image) in which all the voxels are classified. If a regular convolutional network with pooling layers and dense layers was used, the CNN network would lose the “where” information and only retain the “what” information which is not acceptable for image segmentation. In the instance of image segmentation, both “what” as well as “where” information are used. Thus, the image is upsampled to convert a low resolution image to a high resolution image to recover the “where” information. Transposed convolution represented by the white arrow pointing up is an exemplary upsampling technic that may be used in the expansive path 310 for upsampling of the feature map and expanding the size of images.
[0112] After the transposed convolution at block 325, the image is upsized, e.g., from 28×28×1024.fwdarw.56×56×512 via up-convolution (upsampling operators) of 2×2×2 by strides of two in each dimension, and then, the image is concatenated with the corresponding image from the contracting path (see the horizontal gray bar 330 from the contracting path 305) and together makes an image of e.g., size 56×56×1024. The reason for the concatenation is to combine the information from the previous layers (i.e., the high-resolution features from the contracting path 305 are combined with the upsampled output from the expansive path 310) in order to get a more precise prediction. This process continues as a sequence of up-convolutions that halves the number of channels, concatenations with a correspondingly cropped feature map from the contracting path 305, repeated application of convolutions (e.g., two 3×3×3 convolutions) that are each followed by a rectified linear unit (ReLU), and a final convolution in block 335 (e.g., one 1×1×1 convolution) to generate a multi-channel segmentation as a three-dimensional segmentation mask. In order to localize, the U-Net 300 uses the valid part of each convolution without any fully connected layers, i.e., the segmentation map only contains the voxels for which the full context is available in the input image, and uses skip connections that link the context features learned during a contracting block and the localization features learned in an expansion block.
Therefore, in view of Xie, it would have been obvious to one of ordinary skill at the time the invention was made to have the feature of wherein the clustering algorithm is the machine learning model-based clustering algorithm and the machine learning model-based clustering algorithm comprises a neural network that has been trained on training data to receive CDTD data as an input and generate clusters of similar voxels as an output, incorporated in the device of Koay, as modified by Burrus, in order to classify tensors and outputting the classified image, which can decrease the complexity of the image data and focus the machine learning on specific tasks for more efficient use (as stated in Xie ¶ [50]).
Claim(s) 7 and 8 is/are rejected under 35 U.S.C. 103 as being unpatentable over Koay, as modified by Burrus and Xie, as applied to claim 6 above, and further in view of Van Muiswinkel (US Pub 2003/0214289).
Re claim 7: However, Koay fails to specifically teach the features of the method of claim 6, wherein generating the water pool image comprises:
selecting a set of diffusion tensor parameters defining the subset of the plurality of diffusion tensors;
masking the CDTD to select the subset of the plurality of diffusion tensors having parameters that satisfy the set of diffusion tensor parameters; and
weighting voxels in the water pool map based on a number of diffusion tensors in the subset of the plurality of diffusion tensors for each voxel.
However, this is well known in the art as evidenced by Van Muiswinkel. Similar to the primary reference, Van Muiswinkel discloses diffusion tensor MRI (same field of endeavor or reasonably pertinent to the problem).
Van Muiswinkel discloses wherein generating the water pool image comprises:
selecting a set of diffusion tensor parameters defining the subset of the plurality of diffusion tensors (e.g. the eigenvalues and eigenvectors are extracted that are used to define the direction of the tensor, which is taught in ¶ [40]-[46].);
[0040] The DT-MRI imaging data are acquired in a step 154 using imaging sequences such as spin-echo sequences which include additional magnetic field gradient pulses that produce the selected diffusion weighting. Preferably, a multiple-echo sequence is used in which images are acquired with several diffusion weightings 156 corresponding to selected apparent diffusion coefficient (ADC) components of the diffusion coefficient tensor. Six apparent diffusion coefficients are generally sufficient to describe the tensor. In the illustrated embodiment, Six diffusion weightings 156 are collected, with magnetic field gradient pulses applied in the (x, 0, 0), (0, y, 0), (0, 0, z), (x, -y, 0), (x, 0, -z), and (0, y, -z) directions, along with an unweighted image (0, 0, 0). However, other combinations of diffusion weighting can be used instead. Using a multiple-echo sequence advantageously reduces data acquisition time and minimizes motion-induced blurring or misregistration across images. To improve the signal-to-noise ratio, data for a plurality of images are preferably collected for each diffusion weighting. The imaging sequence also optionally includes additional RF pulses or magnetic field gradient pulses or sweeps to compensate for magnetic field gradient-induced eddy currents and other imaging artifacts.
[0041] Image data collected in the step 154 is reconstructed in a step 158 to form diffusion weighted image reconstructions S.sub.o and S.sub.ijk where ijk indicates the various weightings 156. An inverse Fourier transform reconstruction known to the art is suitably used, although other reconstruction methods can also be employed.
[0042] With the diffusion-weighted images acquired and reconstructed, the apparent diffusion coefficients (ADC's) at each voxel are calculated according to equation (2) using linear regression or another technique and the apparent diffusion coefficient tensor map is constructed in a step 162. The eigenvalues and eigenvectors 166 are extracted in a step 164.
[0043] Optionally, an anisotropy map is computed in a step 168, for example in accordance with equations (3) and (4). The anisotropy map is preferably rendered in a step 170, for example by colorizing the voxels based on the anisotropy value, to obtain a colorized anisotropy image 172 for display to an associated user. The anisotropy image 172 provides a convenient medium for the user to select regions of interest for fiber tracking.
[0044] With reference to FIG. 5, a suitable method 190 for tracking fibers in the DT-MRI image is described. In a step 192, the user selects a starting region of interest. In a suitable embodiment, the selection 192 is made with reference to the anisotropy image 172 optionally obtained in the DT-MRI imaging method 150 of FIG. 4. The selected starting region of interest is preferably indicated by the user graphically using a mouse pointer or other graphical selection device. The selected region of interest can be a single voxel, a planar region of voxels, or a three-dimensional region of voxels. Optionally, the user can also select an ending region of interest. Such a selection is typically useful for tracking fibers extending between two functional brain regions of deep white matter.
[0045] A starting voxel within the selected starting region of interest is selected in a step 196. Beginning with this voxel, a local direction is identified in a step 198 corresponding to the major eigenvector e.sub.1 direction (see FIG. 2), for example by ordering the eigenvalues, identifying the largest or major eigenvalue .lambda..sub.1, and identifying the direction of the corresponding major eigenvector e.sub.1. Next voxels are identified in a step 200 which are nearby the current voxel along the local direction (see FIG. 3). In a preferred embodiment, both positive and negative (bi-directional) tracking is performed by identifying next voxels in both positive and negative local diffusion (e.sub.1) directions. As the tracking progresses bi-directionally, a positive fiber end is grown by successively identifying voxels in the positive local direction while a negative fiber end is grown by successively identifying voxels in the negative local direction. Unidirectional fiber tracking is also contemplated for certain situations such as tracking a fiber extending away from a large, dense region of deep white matter.
[0046] In a step 202, a localized interpolation is performed in the vicinity of the current tracking front or fronts, i.e. in the vicinity of the next voxel or voxels. The interpolation is preferably localized to the region around the tracking front (or fronts in bi-directional tracking) to reduce computational time and memory loading. In a preferred embodiment, the interpolation is performed in the tensor while obtaining the eigenvectors and eigenvalues so that the tracking front or fronts progress within a higher resolution local diffusion tensor space. The resultant tracked fiber representation will have an improved resolution and smoothness corresponding to the higher interpolated local resolution of the source tensor map.
[0047] To maintain tracking accuracy in the interpolated higher resolution local diffusion tensor space, the interpolated voxels are preferably weighted by a parameter related to the local anisotropy. For example, the eigenvectors and eigenvalues of an interpolated voxel are suitably selected as an average of the eigenvectors and eigenvalues of nearby voxels so that the interpolated voxel reflects the local anisotropy. Weighting the plurality of eigenvalues and eigenvectors is computationally expensive, however. To improve fiber tracking speed, it has been found to be sufficient to select an interpolated voxel major eigenvector having a direction corresponding to a weighted average of the major eigenvector directions of adjacent voxels. Optionally, the selected major eigenvector includes a magnitude weighting corresponding to the fractional anisotropy or another anisotropy parameter.
masking the CDTD to select the subset of the plurality of diffusion tensors having parameters that satisfy the set of diffusion tensor parameters (e.g. the parameters are used to create a map that contains a masking, or colorized voxels, that are used to satisfy the direction of the eigenvalues and eigenvectors that reflect the direction of water, which is taught in ¶ [42]-[46] above.); and
weighting voxels in the water pool map based on a number of diffusion tensors in the subset of the plurality of diffusion tensors for each voxel (e.g. the voxels are weighted based on the number of tensors resulting from the eigenvalues and eigenvectors magnitude and direction within a selected area. This is used to colorize a map to reflect the diffusion, which is taught in ¶ [43]-[47] above.).
Therefore, in view of Van Muiswinkel, it would have been obvious to one of ordinary skill at the time the invention was made to have the feature of wherein generating the water pool image comprises: selecting a set of diffusion tensor parameters defining the subset of the plurality of diffusion tensors; masking the CDTD to select the subset of the plurality of diffusion tensors having parameters that satisfy the set of diffusion tensor parameters; and weighting voxels in the water pool map based on a number of diffusion tensors in the subset of the plurality of diffusion tensors for each voxel, incorporated in the device of Koay, as modified by Burrus, in order to weight the voxels by a parameter indicative of the diffusion, which produce a display of higher resolution of the acquired diffusion tensor map (as stated in Van Muiswinkel ¶ [07]).
Re claim 8: However, Koay fails to specifically teach the features of the method of claim 7, wherein selecting the set of diffusion parameters comprises defining a cutoff threshold Dcut, and wherein the set of diffusion tensor parameters comprises
PNG
media_image3.png
16
81
media_image3.png
Greyscale
Dcut mm2/s and 3λ1 > λ2, 3λ2 > λ3.
However, this is well known in the art as evidenced by Xie. Similar to the primary reference, Xie discloses classifying an image using a clustering algorithm (same field of endeavor or reasonably pertinent to the problem). Xie discloses wherein selecting the set of diffusion parameters comprises 3λ1 > λ2, 3λ2 > λ3. (e.g. the system discloses the three eigenvalues that are equal to one another or similar. If this is the case, three times the first eigenvalue would be greater than the second eigenvalue, and three times the second eigenvalue would be greater than the third eigenvalue. The different eigenvalues are explained in ¶ [85].)
[0085] In some embodiments, the one or more imaging systems 130 include a DI system (e.g., an MRI system with special software) configured to apply supplemental MR gradients during the image acquisition. The motion of protons during the application of these gradients affects the signal in the images, thereby, providing information on molecular diffusion. A DTI matrix is obtained from a series of diffusion-weighted images in various gradient directions. The three diffusivity parameters or eigenvalues (λ1, λ2, and λ3), are generated by matrix diagonalization. The diffusivities are scalar indices describing water diffusion in a specific voxels (the smallest volumetric elements in the image) associated with the geometry of tissue. Various diffusion imaging techniques may be used to compute diffusion tensor parametric maps and additional image contrasts from the diffusivities. DTI properties or indices represented by these maps may include (but are not limited to) molecular diffusion rate (MD map or ADC map), the directional preference of diffusion (FA map), the AD map (diffusion rate along the main axis of diffusion), and RD map (rate of diffusion in the transverse direction). The diffusivities (λ1, λ2, and λ3) obtained by DTI matrix diagonalization can be delimitated into parallel (λ1) and perpendicular (λ2 and λ3) components to the tissue. The sum of the diffusivities (λ1, λ2, and λ3) is called the trace, while their average (=trace/3) is called the MD or ADC. Fractional anisotropy (FA) is an index for the amount of diffusion asymmetry within a voxel, defined in terms of its diffusivities (λ1, λ2, and λ3). The value of FA varies between 0 and 1. For perfect isotropic diffusion, λ1=λ2=λ3, the diffusion ellipsoid is a sphere, and FA=0. With progressive diffusion anisotropy, the eigenvalues become more unequal, the ellipsoid becomes more elongated, and the FA.fwdarw.1. Axial diffusivity (AD), λ∥≡λ1>λ2, λ3, describes the mean diffusion coefficient of water molecules diffusing parallel to a tract within the voxel of interest. Similarly, radial diffusivity (RD), λ.sup.⊥≡(λ2+λ3)/2, can be defined as the magnitude of water diffusion perpendicular to the main eigenvector.
Therefore, in view of Xie, it would have been obvious to one of ordinary skill at the time the invention was made to have the feature of wherein selecting the set of diffusion parameters comprises 3λ1 > λ2, 3λ2 > λ3, incorporated in the device of Koay, as modified by Burrus and , in order to classify tensors and outputting the classified image, which can decrease the complexity of the image data and focus the machine learning on specific tasks for more efficient use (as stated in Xie ¶ [50]).
However, the combination above fails to specifically teach the features of defining a cutoff threshold Dcut, and wherein the set of diffusion tensor parameters comprises
PNG
media_image3.png
16
81
media_image3.png
Greyscale
Dcut mm2/s.
However, this is well known in the art as evidenced by Van Muiswinkel. Similar to the primary reference, Van Muiswinkel discloses diffusion tensor MRI (same field of endeavor or reasonably pertinent to the problem).
Van Muiswinkel discloses defining a cutoff threshold Dcut, and wherein the set of diffusion tensor parameters comprises
PNG
media_image3.png
16
81
media_image3.png
Greyscale
Dcut mm2/s (e.g. the system discloses a threshold that is set where the eigenvalues are greater when considering the smallest value eigenvalue. The threshold and eigenvalues are explained in ¶ [29]-[31] and [49].).
[0029] With continuing reference to FIG. 1 and with further reference to FIG. 2, an eigenvector/eigenvalue ordering processor 46 orders the diffusion tensor eigenvectors and eigenvalues at voxels. As seen in FIG. 2, the ordered eigenvalues .lambda..sub.1, .lambda..sub.2, .lambda..sub.3 (ordered from largest to smallest eigenvalue) and the corresponding eigenvectors e.sub.1, e.sub.2, e.sub.3 of the diffusion tensor have useful physical significance. The largest eigenvalue is designated in FIG. 2 as .lambda..sub.1. The corresponding eigenvector e.sub.1 is called the major eigenvector, and aligns with the spatial direction having the highest diffusion coefficient. The remaining eigenvalues .lambda..sub.2, .lambda..sub.3 have corresponding eigenvectors e.sub.2, e.sub.3 called the medium and minor eigenvectors in FIG. 2. These eigenvectors e.sub.2, e.sub.3 are orthogonal to e.sub.1 and align with spatial directions having lower diffusion coefficients. The relative values of the eigenvalues .lambda..sub.1, .lambda..sub.2, .lambda..sub.3 are indicative of the spatial orientation and magnitude of the diffusion tensor anisotropy.
[0030] With continuing reference to FIG. 2, the eigenvectors and eigenvalues are geometrically representable by an ellipsoid 100 whose long axis aligns with eigenvector e.sub.1, i.e. with the direction of the highest apparent diffusion coefficient. The deviation of the ellipsoid 100 from a perfect sphere is representative of the anisotropy of the diffusion tensor. An anisotropic diffusion coefficient tensor can reflect the influence of neural fiber bundles 102 which tend to inhibit diffusion in directions partially or totally orthogonal to the fibers 102, e.g. the directions of eigenvectors e.sub.2, e.sub.3. In contrast, diffusion parallel to the fibers 102, i.e. channeled along the direction of the major eigenvector e.sub.1, is enhanced and larger than along the e.sub.2, e.sub.3 directions.
[0031] With returning reference to FIG. 1, an anisotropy map 50, such as a fractional anisotropy map known to the art, or another anisotropy image map that emphasizes the anisotropy magnitude, is optionally calculated from the ordered eigenvectors and eigenvalues. In a suitable embodiment, an anisotropy map is calculated on a per voxel basis according to:
PNG
media_image4.png
258
614
media_image4.png
Greyscale
[0049] The method 190 iteratively repeats the steps 198, 200, 202 to grow the tracked fiber either uni-directionally or bi-directionally. Preferably, a decision step 204 within the iterative loop checks for a termination of a progressing fiber end. One suitable fiber termination criterion includes a fractional anisotropy or other anisotropy magnitude parameter below a selected value, e.g. at or below the FA=0.10 threshold used in equation (3). Since a low anisotropy corresponds with a highly isotropic diffusion tensor, it is reasonable to associate an anisotropy parameter value that drops below a selected threshold with the terminal point of a tracked fiber.
Therefore, in view of Van Muiswinkel, it would have been obvious to one of ordinary skill at the time the invention was made to have the feature of defining a cutoff threshold Dcut, and wherein the set of diffusion tensor parameters comprises
PNG
media_image3.png
16
81
media_image3.png
Greyscale
Dcut mm2/s, incorporated in the device of Koay, as modified by Burrus, in order to set thresholds and a limit to tensor parameters, which can aid in improving diffusion fiber tracking (as stated in Van Muiswinkel ¶ [32]).
Claim(s) 9 is/are rejected under 35 U.S.C. 103 as being unpatentable over Koay, as modified by Burrus, Xie and Van Muiswinkel, as applied to claim 7 above, and further in view of Hardi (NPL High Angular Resolution Diffusion Imaging (pub year: 2015)).
Re claim 9: However, Koay fails to specifically teach the features of the method of claim 7, wherein selecting the set of diffusion parameters comprises defining a cutoff threshold Dcut, and wherein the set of diffusion tensor parameters comprises , λ1 < Dcut, λ2 < Dcut, and (λ1 + λ2) <
PNG
media_image5.png
28
14
media_image5.png
Greyscale
.
However, this is well known in the art as evidenced by Van Muiswinkel. Similar to the primary reference, Van Muiswinkel discloses diffusion tensor MRI (same field of endeavor or reasonably pertinent to the problem).
Van Muiswinkel discloses wherein selecting the set of diffusion parameters comprises defining a cutoff threshold Dcut (e.g. the system discloses a threshold that is set where the eigenvalues are greater when considering the smallest value eigenvalue. The threshold and eigenvalues are explained in ¶ [29]-[31] and [49] above.).
Therefore, in view of Van Muiswinkel, it would have been obvious to one of ordinary skill at the time the invention was made to have the feature of wherein selecting the set of diffusion parameters comprises defining a cutoff threshold Dcut, incorporated in the device of Koay, as modified by Burrus, in order to set thresholds and a limit to tensor parameters, which can aid in improving diffusion fiber tracking (as stated in Van Muiswinkel ¶ [32]).
However, the combination above fails to specifically teach the features of wherein the set of diffusion tensor parameters comprises , λ1 < Dcut, λ2 < Dcut, and (λ1 + λ2) <
PNG
media_image5.png
28
14
media_image5.png
Greyscale
.
However, this is well known in the art as evidenced by Hardi. Similar to the primary reference, Hardi discloses diffusion imaging (same field of endeavor or reasonably pertinent to the problem).
Hardi discloses wherein the set of diffusion tensor parameters comprises , λ1 < Dcut, λ2 < Dcut, and (λ1 + λ2) <
PNG
media_image5.png
28
14
media_image5.png
Greyscale
(e.g. the system can have diffusion parameters that reflect a prolate tensor that includes a tensor that is larger than the other two eigenvalues that are set as low values. With the later eigenvalues set to be lower than a specific minimum value that ensures the needle-like shaped of the tensor image, this performs the feature of having a tensor set to have one very large value for an eigenvalue while the other eigenvalues are set to very low. This is illustrated in figure 5 and explained on pages 4 and 5 of the publication. With the initial reference able to set a cutoff of what the eigenvalues should be above or below combined with this reference, the feature of the claims is disclosed.).
Therefore, in view of Hardi, it would have been obvious to one of ordinary skill before the effective filing date of the claimed invention was made to have the feature of wherein the set of diffusion tensor parameters comprises , λ1 < Dcut, λ2 < Dcut, and (λ1 + λ2) <
PNG
media_image5.png
28
14
media_image5.png
Greyscale
, incorporated in the device of Koay, as modified by Van Muiswinkel, in order to certain set values low for eigenvalues and a higher eigenvalue for a prolate image, which can aid in imaging tissue properties of the tensor (as stated in Hardi page 5).
Claim(s) 9 is/are rejected under 35 U.S.C. 103 as being unpatentable over Koay, as modified by Burrus, Xie and Van Muiswinkel, as applied to claim 7 above, and further in view of Skare (NPL Optimisation Strategies in Diffusion Tensor MR Imaging (pub year: 2002)).
Re claim 10: However, Koay fails to specifically teach the features of the method of claim 7, wherein selecting the set of diffusion parameters comprises defining a cutoff threshold Dcut, and wherein the set of diffusion tensor parameters comprises λ1 < Dcut, λ2 >
PNG
media_image6.png
28
14
media_image6.png
Greyscale
, and λ2< 3λ3.
However, this is well known in the art as evidenced by Van Muiswinkel. Similar to the primary reference, Van Muiswinkel discloses diffusion tensor MRI (same field of endeavor or reasonably pertinent to the problem).
Van Muiswinkel discloses wherein selecting the set of diffusion parameters comprises defining a cutoff threshold Dcut (e.g. the system discloses a threshold that is set where the eigenvalues are greater when considering the smallest value eigenvalue. The threshold and eigenvalues are explained in ¶ [29]-[31] and [49] above.).
Therefore, in view of Van Muiswinkel, it would have been obvious to one of ordinary skill at the time the invention was made to have the feature of wherein selecting the set of diffusion parameters comprises defining a cutoff threshold Dcut, incorporated in the device of Koay, as modified by Burrus, in order to set thresholds and a limit to tensor parameters, which can aid in improving diffusion fiber tracking (as stated in Van Muiswinkel ¶ [32]).
However, this is well known in the art as evidenced by Skare. Similar to the primary reference, Skare discloses diffusion imaging (same field of endeavor or reasonably pertinent to the problem).
Skare discloses wherein the set of diffusion tensor parameters comprises λ1 < Dcut, λ2 >
PNG
media_image6.png
28
14
media_image6.png
Greyscale
, and λ2< 3λ3 (e.g. the publication discloses an eigenvalue that is below a certain number while the other eigenvalues are close in value where a third of one eigenvalue would make that less than the eigenvalue close to the divided eigenvalue’s original value. In addition, if the eigenvalues with similar values has one that is multiplied by 3, that value would be greater than the eigenvalue similar to the original value of the other eigenvalue. This scenario highlights a disc shaped ellipsoid or oblate description, which is taught on page 21 in the skewness section. This feature combined with the setting of the threshold within the prior reference performs the feature of the claim.).
Therefore, in view of Skare, it would have been obvious to one of ordinary skill before the effective filing date of the claimed invention was made to have the feature of wherein the set of diffusion tensor parameters comprises λ1 < Dcut, λ2 >
PNG
media_image6.png
28
14
media_image6.png
Greyscale
, and λ2< 3λ3, incorporated in the device of Koay, as modified by , in order to set parameters associated with eigenvalues and a threshold, which aid in defining the oblate or disc-shaped ellipsoid diffusion (as stated in Skare ¶ [21]).
Claim(s) 15 and 16 is/are rejected under 35 U.S.C. 103 as being unpatentable over Koay, as modified by Burrus, as applied to claim 1 above, and further in view of Ozarslan (US Pub 2023/0124954)
Re claim 15: However, Koay fails to specifically teach the features of the method of claim 1, wherein outputting the CDTD comprises: analyzing the CDTD, using the computer system, to assess different types of diffusion in tissues in the region-of-interest; and generating, with the computer system, a report based on analyzing the CDTD, wherein the report indicates microstructure of the tissues in the region-of-interest.
However, this is well known in the art as evidenced by Ozarslan. Similar to the primary reference, Ozarslan discloses plotting different tensors (same field of endeavor or reasonably pertinent to the problem).
Ozarslan discloses wherein outputting the CDTD comprises: analyzing the CDTD, using the computer system, to assess different types of diffusion in tissues in the region-of-interest; and generating, with the computer system, a report based on analyzing the CDTD, wherein the report indicates microstructure of the tissues in the region-of-interest (e.g. the diffusion data associated with different diffusion gradients are plotted and analyzed. The diffusion propagators are displayed on a screen, which is taught in ¶ [94]-[97]. The data measured in order to display the diffusion can be used to indicate the microstructure of tissues of the data, which is taught in ¶ [96], [97] and [107]-[109].).
[0094] A computer-implemented diffusion magnetic resonance (diffusion-MR) system and method examines specimens such as materials, porous media, food products, and biological tissue by detecting diffusion in the specimen. A pulse sequence featuring gradients of different durations applied at different times employed with an arrayed plurality of different diffusion gradient durations, strengths, and directions collectively provide a Fourier transform of the diffusion propagator. MR diffusion data is collected from the specimen. The diffusion propagator is computed by transforming the data on a computer. The data is used to estimate the steady state particle distribution and other measures. The calculated propagator and measures can be used to create new contrasts in MR imaging that can detect changes in the specimen's microstructure and improve the sensitivity and specificity of MR studies for clinical applications.
[0095] An object of the present invention is to quantify the diffusive motion of the particles. A further object of the present invention is to provide new NMR and MRI techniques that use information contained in the measured diffusion and characterize its evolution in time.
[0096] A method and system for measuring the diffusion propagator of spin-labelled particles is described herein. A pulse sequence is applied using an NMR spectrometer or imager to a sample within the NMR apparatus for generating NMR signals from which the diffusion propagator and related quantities can be calculated. The MR diffusion data are preferably acquired from the tissue at an arrayed plurality of pulse sequences with different diffusion gradient durations, strengths, and/or directions. A computer-implemented diffusion magnetic resonance (diffusion-MR) method analyses a specimen such as a material or porous medium or biological tissue by detecting diffusion in the specimen.
[0097] Calculated MR characteristics are preferably visualized on multi-dimensional plots where spatial coordinates are displayed on different axes. At least one quantity describing the diffusion propagator is calculated using the computer from the MR diffusion data. The estimated propagator may be related to the microstructure of the medium. Spatial images of such quantities describing the underlying diffusion process can be produced and can be related to other structural characteristics within each voxel. Diffusion data obtained using measurements performed with different diffusion gradients are plotted and analysed. The propagators are displayed with contour lines or surfaces. Diffusion measurements may also be employed in calibrating the NMR apparatus itself.
[0107] The methods of the invention include MRI diffusion-weighted imaging (DWI). Briefly, this approach is based on the measurement of random motion of molecules and the capability of nuclear magnetic resonance to quantify diffusional movement of particles. Diffusion imaging is a method that combines this diffusion measurement with MRI. This technique can characterize diffusion properties of spin-labelled molecules at each picture element (pixel or voxel) of an image. Properties of the diffusion of spin labelled molecules are related to the chemical and geometrical environments. For example, diffusion imaging can be used to infer information regarding the microstructure (e.g., cellular membranes and large macromolecules) that restrict or hinder the water molecular motion. Consequently, diffusion imaging can detect water diffusion in highly ordered organs, such as brains. In these tissues, water does not diffuse equally in all directions due to restrictions of water molecules imposed by cellular membranes, resulting in a property called anisotropic diffusion.
[0108] By considering MR data obtained with special gradient pulse sequences, it has been discovered that diffusion within the specimen can be characterized by estimating the diffusion propagator. The obtained diffusion propagator can represent different phenomena. For example, in the embodiment illustrated in FIG. 5, when the first pulse is very long and the second and third pulses are very short, the obtained estimate of the diffusion propagator provides an accurate representation of the real diffusion propagator. When the pulse durations differ from the said conditions, the invention can still be employed and the obtained quantity represents an apparent diffusion propagator, which may not represent the true diffusion propagator. For example, when the durations of the second and third pulses in FIG. 5 are finite, the arguments of the estimated diffusion propagator, x and x′, represent the particle positions averaged over the duration of the gradients.
[0109] This new method for measuring diffusion provides previously undiscovered information about the specimen that can be used to generate data and images from the data and obtain new contrasts based on different mathematical parameters. These new contrast mechanisms should provide additional information about the material or tissue microstructure that can improve the specificity and utility of diffusion MR for characterizing alterations, e.g. due to disease in human patients. The method can be practiced in vitro, ex vivo or in vivo.
Therefore, in view of Ozarslan, it would have been obvious to one of ordinary skill at the time the invention was made to have the feature of wherein outputting the CDTD comprises: analyzing the CDTD, using the computer system, to assess different types of diffusion in tissues in the region-of-interest; and generating, with the computer system, a report based on analyzing the CDTD, wherein the report indicates microstructure of the tissues in the region-of-interest, incorporated in the device of Koay, in order to determine the types of diffusion and indicate the diffusion in an output, which can improve the determination of diffusion parameters for a specimen using MRI (as stated in Ozarslan ¶ [06]).
Re claim 16: However, Koay fails to specifically teach the features of the method of claim 15, wherein the report indicates a heterogeneity of the tissues in the region-of-interest.
However, this is well known in the art as evidenced by Ozarslan. Similar to the primary reference, Ozarslan discloses plotting different tensors (same field of endeavor or reasonably pertinent to the problem).
Ozarslan discloses wherein the report indicates a heterogeneity of the tissues in the region-of-interest (e.g. the heterogeneity can be indicated through the plots of the tissues related to the areas of interests, which is taught in ¶ [89], [90], [128] and [130].).
[0089] FIG. 10a-d depict schematically a specimen comprising five circular pores and quantities related to the heterogeneity of the specimen. FIG. 10a depicts the estimation of the dispersity index for a specimen comprising 5 circular pores of radii 12, 36, 60, 84, and 108 units. FIG. 10b depicts the same pores when their centres coincide. The technique effectively performs the measurement on the system depicted in FIG. 10b. The distance from the common centre in
[0090] FIG. 10b is r. FIGS. 10c and 10d plot two quantities that can be obtained using the proposed method against r. The r value that maximizes the first quantity is employed to estimate the dispersion index as shown in FIG. 10d.
[0128] Another example involves a structurally heterogeneous specimen comprising N pores. In this case, the steady state distribution of particles in nth pore can be denoted by ρ.sub.n(x). Furthermore, ƒ.sub.n denotes the signal fraction contributed by the nth pore to the signal when the gradients are turned off. The d-dimensional inverse Fourier transform of E.sub.Δ(q, 0) or E.sub.Δ(0, q′) is
[0130] A quantity, which can be referred to as “local variance” can be defined through σ.sup.2(x)=μ.sub.2(x)−ρ(x).sup.2. The square root of this quantity (“local standard deviation”) can be made dimensionless through σ.sub.l(x)=σ(x)/ρ(x), or via σ.sub.g(x)=σ(x)/ρ(x.sub.0), where x.sub.0 can be taken to be any point with no vanishing ρ(x.sub.0). A particular choice of x.sub.o0 is the one that maximizes ρ(x.sub.0). These maps are expected to be descriptive of the space-dependent heterogeneity of the specimen when the centers-of-mass of all pores are brought to the same point.
Therefore, in view of Ozarslan, it would have been obvious to one of ordinary skill at the time the invention was made to have the feature of wherein the report indicates a heterogeneity of the tissues in the region-of-interest, incorporated in the device of Koay, in order to determine the types of diffusion and indicate the diffusion in an output, which can improve the determination of diffusion parameters for a specimen using MRI (as stated in Ozarslan ¶ [06]).
Claim(s) 17 and 18 is/are rejected under 35 U.S.C. 103 as being unpatentable over Koay, as modified by Burrus and Ozarslan, as applied to claim 15 above, and further in view of Xie.
Re claim 17: However, Koay fails to specifically teach the features of the method of claim 15, wherein analyzing the CDTD comprises generating a plurality of water pool images from the CDTD and outputting the plurality of water pool images as part of the report, wherein each of the plurality of water pool images depicts water diffusion associated with a different subset of the plurality of diffusion tensors.
However, this is well known in the art as evidenced by Xie. Similar to the primary reference, Xie discloses classifying an image using a clustering algorithm (same field of endeavor or reasonably pertinent to the problem). Xie discloses wherein analyzing the CDTD comprises generating a plurality of water pool images from the CDTD and outputting the plurality of water pool images as part of the report, wherein each of the plurality of water pool images depicts water diffusion associated with a different subset of the plurality of diffusion tensors (e.g. image comprised of the diffused degree of water is generated and can be displayed, which is taught in ¶ [63], [66] and [125] above. The different diffusion depicted in the display is related to the different degree of diffusion associated with the tensors. Different colors can be represented within the display.).
Therefore, in view of Xie, it would have been obvious to one of ordinary skill at the time the invention was made to have the feature of wherein analyzing the CDTD comprises generating a plurality of water pool images from the CDTD and outputting the plurality of water pool images as part of the report, wherein each of the plurality of water pool images depicts water diffusion associated with a different subset of the plurality of diffusion tensors, incorporated in the device of Koay, as modified by Burrus, in order to classify tensors and outputting the classified image, which can decrease the complexity of the image data and focus the machine learning on specific tasks for more efficient use (as stated in Xie ¶ [50]).
Re claim 18: However, Koay fails to specifically teach the features of the method of claim 15, wherein analyzing the CDTD comprises generating a classification image from the CDTD and outputting the classification image as part of the report, wherein the classification image depicts classified regions in the region-of-interest that are classified based on subsets of the plurality of diffusion tensors.
However, this is well known in the art as evidenced by Xie. Similar to the primary reference, Xie discloses classifying an image using a clustering algorithm (same field of endeavor or reasonably pertinent to the problem). Xie discloses wherein analyzing the CDTD comprises generating a classification image from the CDTD and outputting the classification image as part of the report, wherein the classification image depicts classified regions in the region-of-interest that are classified based on subsets of the plurality of diffusion tensors (e.g. the invention discloses an output of a classified voxel that is acquired from the tensors that have been segmented and classified. The region of interest includes the classified image data that corresponds to the tensors processed, which is taught in ¶ [81], [82] and [125] above. Combining this output with the above plots in the previously applied references can output the region of interests that are classified.).
Therefore, in view of Xie, it would have been obvious to one of ordinary skill at the time the invention was made to have the feature of wherein analyzing the CDTD comprises generating a classification image from the CDTD and outputting the classification image as part of the report, wherein the classification image depicts classified regions in the region-of-interest that are classified based on subsets of the plurality of diffusion tensors, incorporated in the device of Koay, as modified by Burrus and Ozarslan, in order to classify tensors and outputting the classified image, which can decrease the complexity of the image data and focus the machine learning on specific tasks for more efficient use (as stated in Xie ¶ [50]).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Yon et al discloses a diffusion tensor distribution.
Tuch discloses in chapter 7 the process multiple-tensor imaging where multiple tensors are detected associated with a single voxel.
Wong discloses multiple tensors associated with a single voxel to estimate direction in cross-fiber sections.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to CHAD S DICKERSON whose telephone number is (571)270-1351. The examiner can normally be reached Monday-Friday 10AM-6PM EST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Abderrahim Merouan can be reached at 571-270-5254. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/CHAD DICKERSON/ Primary Examiner, Art Unit 2682
Prosecution Insights