DETAILED ACTIONNotice 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 Arguments
Applicant’s arguments and amendments in the Amendment filed June 5, 2026 (herein “Amendment”), with respect to the objection of claim 39 have been fully considered and are persuasive. The objection of claim 39 has been withdrawn.
Applicant’s arguments and amendments in the Amendment, with respect to the rejections of claims 1, 12, 9 and 28 and therefore any claims depending therefrom under 35 U.S.C. 112(b) for indefiniteness have been fully considered and are persuasive. The rejections of claims 1, 12, 9 and 28 and therefore any claims depending therefrom under 35 U.S.C. 112(b) have been withdrawn.
Applicant’s arguments and amendments with respect to the rejection(s) of claim(s) 1 and 12 and various claims depending therefrom under 35 U.S.C. 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 Therefore, the rejection has been withdrawn.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 Principe et al., US Patent No. 11,721,519 B2.
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 following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
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.
Claims 1–9, 11–14, 16–24, 26, 28–36, and 40 are rejected under 35 U.S.C. 103 as being unpatentable over Mckinnon et al., US Patent No. 12,510,614 B2 (herein “Mckinnon”) in view of Zbontar, Jure, et al. "fastMRI: An open dataset and benchmarks for accelerated MRI." arXiv preprint arXiv:1811.08839 (2018), cited in the IDS filed 9/5/2025 (herein “Zbontar”) in view of Principe et al., US Patent No. 11,721,519 B2 (herein “Principe”).
Regarding claims 1 and 12, with substantive differences between the claims noted in curly brackets {}, deficiencies of Mckinnon noted in square brackets [], and claim 1 as exemplary, Mckinnon teaches a {method comprising / system comprising a first artificial intelligence (AI) engine configured to} (Mckinnon Abstract, col. 16, l. 64–col. 17, l. 40, systems and methods for reconstructing MRI images, including a deep learning model including a convolutional neural network (CNN)):
receiving, by a first artificial intelligence (AI) engine, a plurality of incomplete magnetic resonance (MR) K-space data [matrices] (McKinnon col. 7, l. 59 – col. 8, l. 4, col. 8, ll. 43–56, and col. 18, ll. 24–34, image processing device acquires image data from an MRI apparatus, the image data being multi-coil undersampled (incomplete) k-space data) of an object scanned by an MR device (Mckinnon col. 7, ll. 41–42, during an MRI scan, a subject (object) is positioned within the imaging space and MR signals of the subject are obtained),
the MR device using [a plurality] of sparse-sampled MR scans, each of the sparse-sampled MR scans of the plurality of sparse-sampled MR scans employing a sparse-sampled MR scan acquisition sequence [having a unique sampling pattern] (Examiner Note: claims 1 and 12 do not positively recite as an element of the method claim the “MR device” but rather the MR device is recited as the source of the scan data, where the scan data itself is not claimed as being generated or created, rather it is “received” (with the generation of the scans by the MR device, and the MR device itself, being outside the scope of this claim) accordingly, appropriate patentable weight is given including considerations of obviousness in combining references to teach a particular scan configuration that is not, again, being positively claimed as being generated by the claims of the present application)(Mckinnon col. 13, ll. 34–55, undersampled data is acquired using accelerated MRI techniques where fewer k-space samples are acquired to reduce scan time (sparse-sampled MR scan acquisition sequence), according to an undersampling pattern, a pre-determined pattern according to which k-space is sampled via a plurality of receive coils of the MRI device, where the undersampling pattern includes a higher sampling density near a k-space origin and a reduced sampling density in higher frequency regions of k-space),
wherein each of the plurality of incomplete MR K-space data [matrices] comprises complex values and is the result of [a different one of the plurality of] spare-sampled MR scans; (Mckinnon col. 18, ll. 28–34, complex data including phase and magnitude information obtained from the acquired multi-coil undersampled k-space data) and
reconstructing, by the first AI engine, a complete MR K-space data matrix of the scanned object corresponding to a complete MR K-space acquisition (Mckinnon col. 13, ll. 32–39, fig. 5, col. 14, ll. 52 – col. 15, ll. 18, reconstructing an MRI image (an image is a matrix of pixel values) from multi-coil undersampled k-space data including producing a data-consistent synthetical multi-coil fully sampled (complete) k-space data, using a trained deep learning regularizer (first AI engine)), wherein the reconstruction is based on the data in the plurality of incomplete MR K-space data [matrices] (Mckinnon col. 14, l. 61–col. 15, l. , data-consistent synthetic multi-coil fully sampled k-space data is combined with (thus based on) the multi-coiled undersampled (incomplete) k-space data to produce an updated estimate that is inverse Fourier transformed to eventually produce the reconstructed MRI image).
Although Mckinnon teaches incomplete MR K-space data and complete MR K-space data, Mckinnon does not explicitly teach these to be matrices.
Further, while as noted above in view of the appropriate patentable weight for the limitations directed towards the format of the data/scans received (and not requiring generation of same), while the system of McKinnon is capable of receiving a plurality of sparse-sampled MR scans, each one having a unique sampling pattern in the scan sequence, nonetheless, McKinnon does not explicitly recite that the sparse-sampled MR scans it receives are specifically of the format claimed: the plurality of sparse-sampled scans employing a sparse-sampled scan acquisition sequence having a unique sampling pattern” and “a different one of the plurality of spare-sampled scans.”
Zbontar teaches incomplete MR K-space data matrices (Zbontar section 2.1, multiple receiver coils produce a separate k-space measurement matrix).
Principe teaches the plurality of sparse-sampled scans employing a sparse-sampled scan acquisition sequence having a unique sampling pattern (Principe col. 2, ll. 37–60, col. 3, ll. 18–65, col. 14, ll. 38–46, sparse sampling to reduce acquisition times in magnetic scan devices, where the sparse sampling approach employs compound signal converters including a primary carrier signal converter modulated by a secondary signal converter which is programmatically randomized (thereby producing a unique sampling pattern), having randomness programmed into the aggregate X-Y pattern, and where the sparse sampling operation is repeated successively over the same area using a uniquely randomized pattern (thus plurality of sparse-sampled scans)), and a different one of the plurality of spare-sampled MR scans (Principe col. 14, ll. 38–46, sampling operation repeated with uniquely randomized pattern for each successive scan, thus each processing of the scans will be on a different one of the scans).
Therefore, taking the teachings of Mckinnon and Zbontar together as a whole, it would have been obvious to a person having ordinary skill in the art (herein “PHOSITA”) before the effective filing date of the claimed invention to have modified the k-space data of Mckinnon to be matrices as disclosed in Zbontar at least as doing so would allow for processing of data using machine learning with compressed sensing which would reduce MR scan time. See Zbontar section 1.
Further, taking the teachings of Mckinnon as modified by Zbontar and Principe together as a whole, it would have been obvious to a PHOSITA before the effective filing date of the claimed invention to have modified the k-space data of Mckinnon to be from a plurality of scans with sparse sampling from a scan sequence having a unique pattern as disclosed in Principe at least as doing so would allow for higher scanning rates while mitigating serial scanning artifacts thus benefitting the quality of sparse sampling and sparse sampling reconstruction. See Principe col. 2, ll. 61–64.
Regarding claims 2 and 13, with claim 2 as exemplary, Mckinnon teaches wherein each unique sampling pattern is unique across the phase-encoded dimension of K-space, such that the plurality of incomplete MR K-space data [matrices] contains variations in sampled spatial information due to a difference in the unique sampling patterns (Mckinnon col. 7, ll. 41–58, MR signals of the subject are obtained by an acquisition protocol that includes a plurality of pulse sequences where a gradient coil is controlled to spatially encode resultant MR signals (phase-encoded) which are digitized and stored in k-space, and where different correspondence of k-space data to pulse sequence can be configured, resulting in different unique sampling patterns and variations in sampled spatial information). McKinnon does not explicitly teach the MR k-space data to be in matrices form, however, Zbontar teaches this in section 2.1, noted above in the independent claim rejection, and with the same motivation to combine these teachings of Zbontar with McKinnon as given above in the independent claim rejection rationale.
Regarding claims 3 and 17, with claim 3 as exemplary, Mckinnon teaches generating, by the first AI engine, a reconstructed MR image matrix based on the reconstructed complete MR K-space data matrix by performing the inverse Fourier transform on the reconstructed complete MR K-space data matrix (Mckinnon col. 12, l. 58–col. 13, l. 13, missing k-space data points using the undersampling pattern are filled in to produce the ith estimated multi-coil fully-sampled (complete) k-space data, which is then transformed by an inverse Fourier transform to result in the ith or final estimated multi-coil MRI image, where an image is 2D data and thus a matrix).
Regarding claims 4 and 18, with claim 4 as exemplary, and deficiencies of Mckinnon noted in square brackets [], Mckinnon teaches further comprising training the first AI engine using a first AI engine training data generator configured to implement a gradient descent algorithm and a backpropagation algorithm with training data comprising a first plurality of input MR image [matrices] and a plurality of ground truth MR image [matrices], wherein (Mckinnon col. 24, ll. 51–63, method for training the deep learning regularizers which reconstruct the MR images the method starting with selecting a training pair consisting of multi-coil undersampled k-space data and corresponding ground truth multi-coil fully sampled k-space data, and where col. 26, ll. 20–34 teaches the training using backpropagation and stochastic gradient descent algorithms):
each ground truth MR image matrix of the plurality of ground truth MR image [matrices] corresponds to a fully-sampled MR K-space data [matrix] (Mckinnon col. 24, ll. 60–67, in the training pair, the ground truth item is a multi-coil fully-sampled k-space data corresponding to MR images),
each input MR image matrix of the first plurality of input MR image matrices is an image matrix corresponding to a down-sampled version of a MR K-space data [matrix] corresponding to one ground truth MR image matrix of the plurality of ground truth MR image [matrices] (Mckinnon col. 24, ll. 60–67, in the training pair, the other item besides the ground truth item is a multi-coil undersampled (down-sampled version) k-space data that corresponds to the ground truth data), the down-sampling having been performed using a first one of the unique sampling patterns (Mckinnon col. 24, ll. 63–66, training data generated according to methods disclosed therein, where col. 10, ll. 31–38, teaches the undersampling (down-sampling) for the undersampled k-space data according to an undersampling pattern), and
the gradient descent algorithm and the backpropagation algorithm iteratively generates successive pluralities of predicted output MR image matrices based on the first plurality of input MR image matrices (Mckinnon col. 26, ll. 35–38, and col. 25, ll. 58–61, a loss is calculated from the combined ground truth fully-sampled image data (where images are matrices of pixel values) with the combined estimated fully-sampled image data (estimated from the deep learning regularizer), and iterations of this calculated loss are performed), until the output of a loss function of the nth successive plurality of predicted output MR image matrices and the plurality of ground truth MR image matrices is below a value, thereby defining weights and biases to be applied by the first AI engine (Mckinnon col. 25, l. 58–col. 26, l. 38, at operation 1312, the image processing system backpropagates the loss through the plurality of deep learning regularizers using a backpropagation algorithm such as stochastic gradient descent, resulting in a trained set of weights for the deep learning regularizers, and where the parameters of the regularizers are updated in the direction that minimizes the loss within a predetermined number of iterations, where the loss function minimized within these number of iterations is below a value). McKinnon does not explicitly teach the MR k-space data to be in matrices form, however, Zbontar teaches this in section 2.1, noted above in the independent claim rejection, and with the same motivation to combine these teachings of Zbontar with McKinnon as given above in the independent claim rejection rationale.
Regarding claims 5 and 19, with claim 5 as exemplary, Mckinnon teaches wherein the training data comprises a second plurality of input MR image matrices, each input MR image of the second plurality of input MR image matrices is an image matrix corresponding to a down-sampled version of a MR K-space data matrix corresponding to one ground truth MR image matrix of the plurality of ground truth MRI image [matrices] (Mckinnon col. 24, ll. 60–67, in the training pair, the other item besides the ground truth item is a multi-coil undersampled (down-sampled version) k-space data that corresponds to the ground truth data), the down-sampling having been performed using a second one of the unique sampling patterns (Mckinnon col. 24, ll. 63–66, training data generated according to methods disclosed therein, where col. 10, ll. 31–38, teaches the undersampling (down-sampling) for the undersampled k-space data according to an undersampling pattern), where the second one of the unique sampling patterns is different than the first one of the unique sampling patterns (Mckinnon col. 28, ll. 61-67, the undersampling pattern determined according to an acceleration factor range from 3 to 5 for two-dimensional 2D MRI data, thus teaching different patterns according to a different acceleration factor). McKinnon does not explicitly teach the MR k-space data to be in matrices form, however, Zbontar teaches this in section 2.1, noted above in the independent claim rejection, and with the same motivation to combine these teachings of Zbontar with Mckinnon as given above in the independent claim rejection rationale.
Regarding claims 6 and 20, with claim 6 as exemplary, Mckinnon teaches wherein the gradient descent algorithm and the backpropagation algorithm iteratively generates the successive pluralities of predicted output MR image matrices based on the first plurality of input MR image matrices and the second plurality of input MR images (Mckinnon col. 24, ll. 51–63, method for training the deep learning regularizers which reconstruct the MR images the method starting with selecting a training pair consisting of multi-coil undersampled k-space data and corresponding ground truth multi-coil fully sampled k-space data, and where col. 26, ll. 20–34 teaches the training using backpropagation and stochastic gradient descent algorithms).
While Mckinnon does not explicitly teach the model training to be using training sets with the different acceleration factors, simply iterating the disclosed training process in Mckinnon twice to account for different training data would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention as doing so would have merely been a duplication of parts (steps). See MPEP §2144.04(VI)(B).
McKinnon does not explicitly teach the MR k-space data to be in matrices form, however, Zbontar teaches this in section 2.1, noted above in the independent claim rejection, and with the same motivation to combine these teachings of Zbontar with Mckinnon as given above in the independent claim rejection rationale.
Regarding claims 7 and 21, with claim 7 as exemplary, Mckinnon teaches wherein each predicted output MR image matrix is based on a corresponding input MR image matrix from the first plurality of input MR image matrices and a corresponding input MR image matrix from the second plurality of input MR image matrices (Mckinnon col. 25, ll. 5–55, image processing system estimates multi-coil fully sampled k-space data (predicted output) from the multi-coil undersampled k-space data using a plurality of deep learning regularizers, where ground truth fully sampled data is converted to image domain via IFFT resulting in multiple complex coil images with respective sensitivity maps (thus different first and second ground truth training MR images used to train the deep learning regularizers, thus the estimated fully sampled k-space data being based on the corresponding image matrix from first and second plurality of input MR images), where images are matrices).
Regarding claim 8, Mckinnon teaches wherein: the first plurality of input MR image matrices comprises a first plurality of real-valued input MR image matrices and a first plurality of imaginary-valued input MR image matrices, the second plurality of input MR image matrices comprises a second plurality of real-valued input MR image matrices and a second plurality of imaginary-valued input MR image matrices (Mckinnon col. 9, ll. 5–12, images input into the deep learning regularizers are complex and comprised of imaginary components (imaginary-valued MR image matrix) and real components (real-valued MR image matrix), and where images are matrices, and where col. 21, l. 61–col. 22, l. 8 teaches the input training data including corresponding undersampled k-space data acquired from plurality of receive coils of the MRI device, and where col. 21, ll. 6–11 teaches that the k-space data is converted to an MRI image via iFFT, where images are matrices), the plurality of ground truth MR image matrices comprises a plurality of real-valued ground truth MR image matrices and a plurality of imaginary-valued ground truth MR image matrices (Mckinnon col. 9, ll. 5–12, images input into the deep learning regularizers are complex and comprised of imaginary components (imaginary-valued MR image matrix) and real components (real-valued MR image matrix), and where images are matrices, and where col. 21, l. 61–col. 22, l. 8 teaches the input training data including ground-truth multi-coil fully-sampled data acquired from plurality of receive coils of the MRI device, and where col. 21, ll. 6–11 teaches that the k-space data is converted to an MRI image via iFFT, where images are matrices), and the plurality of predicted output MR image matrices comprises a plurality of real-valued predicted output MR image matrices and a plurality of imaginary-valued predicted output MR image matrices (Mckinnon col. 29, ll. 1–8, and 20–25, the trained deep learning regularizer outputting a real-component predicted regularized MRI image and an imaginary component of the predicted regularized MRI image, where there are multiple trained deep learning regularizers with their own respective outputs, thus plurality of predicted output MR images, where images are pixel matrices).
Regarding claims 9 and 28, with claim 9 as exemplary, Mckinnon teaches wherein the resolution of spatial features in each ground truth MR image matrix of the plurality of ground truth MR image matrices is preserved in the corresponding predicted output MR image matrix of the nth successive plurality of predicted output MR image matrices (Mckinnon col. 21, ll. 61–67 and 6–11, ground truth captures k-space data sufficient for high-quality image reconstruction (predicted output from reconstruction preserving ground-truth quality of resolution), where an iFFT is applied to the k-space data to convert it to an image which is a matrix).
Regarding claims 11 and 40, with claim 11 as exemplary, Mckinnon teaches wherein receiving, by the first AI engine, the plurality of incomplete MR K-space data matrices of an object scanned by an MR device comprises receiving the plurality of incomplete MR K-space scans as multi-channel inputs (Mckinnon col. 18, ll. 5–10, the deep learning regularizer architecture is adapted to process the multiple channels of data from the different coils of the multi-coil MRI data (multi-channel inputs)).
Regarding claims 14 and 26, with claim 14 as exemplary, Mckinnon teaches wherein each unique sampling pattern is a unique realization of the same probability distribution function (Mckinnon col. 28, ll. 61–67, the undersampling pattern is a pseudo-random variable-density undersampling pattern, thus having a probability distribution function that is pseudo-random for all (same), but varies (each unique) with an acceleration factor ranging between 3 to 5).
Regarding claim 16, Mckinnon teaches the first AI engine is further configured to generate a reconstructed MR image matrix based on the reconstructed complete MR K-space data matrix (Mckinnon col. 12, l. 58–col. 13, l. 13, missing k-space data points using the undersampling pattern are filled in to produce the ith estimated multi-coil fully-sampled (complete) k-space data, which is then transformed by an inverse Fourier transform to result in the ith or final estimated (reconstructed) multi-coil MRI image, where an image is 2D data and thus a matrix).
Regarding claim 22, Mckinnon teaches wherein: each input MR image matrix of the first plurality of input MR image matrices comprises complex values, each input MR image matrix of the second plurality of input MR image matrices comprises complex values (Mckinnon col. 9, ll. 5–12, images input into the deep learning regularizers are complex and comprised of imaginary components (imaginary-valued MR image matrix) and real components (real-valued MR image matrix), and where images are matrices, and where col. 21, l. 61–col. 22, l. 8 teaches the input training data including corresponding undersampled k-space data acquired from plurality of receive coils of the MRI device, and where col. 21, ll. 6–11 teaches that the k-space data is converted to an MRI image via iFFT, where images are matrices), each ground truth MR image matrix of the plurality of ground truth MR image matrices comprises complex values (Mckinnon col. 9, ll. 5–12, images input into the deep learning regularizers are complex and comprised of imaginary components (imaginary-valued MR image matrix) and real components (real-valued MR image matrix), and where images are matrices, and where col. 21, l. 61–col. 22, l. 8 teaches the input training data including ground-truth multi-coil fully-sampled data acquired from plurality of receive coils of the MRI device, and where col. 21, ll. 6–11 teaches that the k-space data is converted to an MRI image via iFFT, where images are matrices), and each predicted output MR image matrix of the plurality of predicted output MR image matrices comprises complex values (Mckinnon col. 29, ll. 1–8, and 20–25, the trained deep learning regularizer outputting a real-component predicted regularized MRI image and an imaginary component of the predicted regularized MRI image, where there are multiple trained deep learning regularizers with their own respective outputs, thus plurality of predicted output MR images, where images are pixel matrices).
Regarding claim 23, Mckinnon teaches wherein: the first plurality of input MR image matrices comprises a first plurality of real-valued input MR image matrices and a first plurality of imaginary-valued input MR image matrices, the second plurality of input MR image matrices comprises a second plurality of real-valued input MR image matrices and a second plurality of imaginary-valued input MR image matrices (Mckinnon col. 9, ll. 5–12, images input into the deep learning regularizers are complex and comprised of imaginary components (imaginary-valued MR image matrix) and real components (real-valued MR image matrix), and where images are matrices, and where col. 21, l. 61–col. 22, l. 8 teaches the input training data including corresponding undersampled k-space data acquired from plurality of receive coils of the MRI device, and where col. 21, ll. 6–11 teaches that the k-space data is converted to an MRI image via iFFT, where images are matrices), the plurality of ground truth MR image matrices comprises a plurality of real-valued ground truth MR image matrices and a plurality of imaginary-valued ground truth MR image matrices (Mckinnon col. 9, ll. 5–12, images input into the deep learning regularizers are complex and comprised of imaginary components (imaginary-valued MR image matrix) and real components (real-valued MR image matrix), and where images are matrices, and where col. 21, l. 61–col. 22, l. 8 teaches the input training data including ground-truth multi-coil fully-sampled data acquired from plurality of receive coils of the MRI device, and where col. 21, ll. 6–11 teaches that the k-space data is converted to an MRI image via iFFT, where images are matrices), and the plurality of predicted output MR image matrices comprises a plurality of real-valued predicted output MR image matrices and a plurality of imaginary-valued predicted output MR image matrices (Mckinnon col. 29, ll. 1–8, and 20–25, the trained deep learning regularizer outputting a real-component predicted regularized MRI image and an imaginary component of the predicted regularized MRI image, where there are multiple trained deep learning regularizers with their own respective outputs, thus plurality of predicted output MR images, where images are pixel matrices).
Regarding claim 24, Mckinnon teaches wherein the first AI engine training data generator is further configured to: generate the plurality of ground truth MR image matrices by performing the inverse Fourier transform on a plurality of fully-sampled MR K-space data matrices (Mckinnon col. 25, ll. 18–29, ground truth image data (images are matrices) is yielded (generate) from a transform of ground-truth multi-coil fully sampled k-space data to an image domain using an inverse Fourier transform IFFT); and generate the first plurality of input MR image matrices by: down-sampling each fully-sampled MR K-space data matrix of the plurality of fully-sampled MR K-space data matrices using the unique sampling pattern; and performing the inverse Fourier transform on each of the plurality of down-sampled MR K-space data matrices (Mckinnon col. 18, ll. 24–34, input to the deep learning regularizer is a multi-coil complex MRI image that is the result of an inverse Fourier transform applied to initially acquired multi-coil undersampled k-space data, where col. 7, ll. 41–58 and col. 8, ll. 43–50, teach the acquisition protocol digitizing and storing the data in k-space according to a sampling of the raw MR signals into k-space, resulting in an undersampled (downsampled) k-space).
Regarding claim 29¸ Mckinnon teaches wherein for each predicted output MR image matrix, the first AI engine is further configured to: transform said predicted output MR image matrix into an equivalent complete MR K-space data matrix (Mckinnon col. 25, ll. 45–62, the estimated (predicted) fully-sampled image domain value is compared with a ground truth value in the k-space domain, where col. 16, ll. 12–14 teaches using a Fourier transform to convert image domain data into the k-space domain);
enforce a data consistency constraint by overriding values in said equivalent complete MR K-space data matrix with the corresponding values in the corresponding fully-sampled MR K-space data matrix, thereby generating a data consistent equivalent complete MR K-space data matrix (Mckinnon col. 25, l. 58–col. 26, l. 38, a loss (data consistency constraint) between the comparison of the estimated fully-sampled k-space data (in k-space domain) and the ground truth fully-sampled k-space image data (corresponding fully-sampled MR K-space data matrix) is minimized (enforced) over multiple iterations, resulting in an optimally trained deep learning regularizer which outputs loss minimized (data consistent) estimated fully-sampled k-space data, where col. 10, ll. 59–61 teaches that the iterative optimization includes constraints that enforce consistency of the k-space kernel with the known properties of the MRI system); and transform said data consistent equivalent complete MR K-space data matrix into an updated predicted output MR image matrix (Mckinnon col. 25, ll. 45–48, and col. 26, ll. 35–43, estimated multi-coil fully sampled k-space data is transformed to the image domain using an IFFT, where the deep learning regularizers output for a particular iteration within a pre-determined number of iterations, where a particular iteration would output the updated estimated (predicted) fully sampled k-space data, then convert to an image using IFFT).
Regarding claim 30, McKinnon teaches wherein the first AI engine is further configured to: transform said predicted output MR image into the equivalent complete MR K-space data matrix comprises performing the Fourier transform on the predicted output MR image (Mckinnon col. 25, ll. 45–62, the estimated (predicted) fully-sampled image domain value is compared with a ground truth value in the k-space domain, where col. 16, ll. 12–14 teaches using a Fourier transform to convert image domain data into the k-space domain); and transform said data consistent equivalent complete MR K-space data matrix into an updated predicted output MR image matrix comprises performing the inverse Fourier transform on the data consistent equivalent complete MR K-space data matrix (Mckinnon col. 25, ll. 45–48, and col. 26, ll. 35–43, estimated multi-coil fully sampled k-space data is transformed to the image domain using an IFFT, where the deep learning regularizers output for a particular iteration within a pre-determined number of iterations, where a particular iteration would output the updated estimated (predicted) fully sampled k-space data, then convert to an image using IFFT).
Regarding claim 31, Mckinnon teaches further comprising an MR device configured to generate the plurality of incomplete MR K-space data matrices by: implementing on the MR device, said sparse-sampled MR scan acquisition sequence, thereby generating MR signal data at a receiver coil of the MR device (Mckinnon col. 7, ll. 41–50, acquisition protocol during an MRI scan using an MRI apparatus including receipt of pulse sequences controlled to spatially encode received by the RF coil unit resulting in raw MR signals); sampling the MR signal data (Mckinnon col. 7, ll. 49–58, one line of k-space data is filled with the raw MR signals per pulse sequence as a repetition time (sampling), also the MR signals received by the coil are digitized before being stored in k-space, where digitization is another kind of sampling); and storing the sampled MR signal data as a MR K-space data [matrix], thereby generating one of the plurality of incomplete MR K-space data matrices (Mckinnon col. 8, l. 43–46, a processor receives the multi-coil undersampled k-space data from the MRI apparatus, where a processor has a cache memory or buffer to receive data for processing, and thus stores the received data therein). McKinnon does not explicitly teach the MR k-space data to be in matrices form, however, Zbontar teaches this in section 2.1, noted above in the independent claim rejection, and with the same motivation to combine these teachings of Zbontar with Mckinnon as given above in the independent claim rejection rationale.
Regarding claim 32, Mckinnon teaches wherein each of the plurality of incomplete MR K-space data matrices and the complete MR K-space data matrix is a two-dimensional MR K-space data matrix (Mckinnon col. 19, ll. 54–56, the multi-coil fully sampled (complete) k-space MRI data is converted to 2D data, and col. 21, ll. 31–34 teaching undersampled (incomplete) MRI k-space data being 2D data). McKinnon does not explicitly teach the MR k-space data to be in matrices form, however, Zbontar teaches this in section 2.1, noted above in the independent claim rejection, and with the same motivation to combine these teachings of Zbontar with Mckinnon as given above in the independent claim rejection rationale.
Regarding claim 33, Mckinnon teaches wherein the first AI engine is a convolutional neural network based on a first deep learning model (Mckinnon col. 22, ll. 52–59, the deep learning regularizer including a neural network model comprising a convolutional neural network CNN architecture).
Regarding claim 34, Mckinnon teaches further comprising a second AI engine configured to select the unique sampling pattern for each sparse-sampled MR scan acquisition sequence (Mckinnon col. 10, l. 39 – col. 11, l. 21, a k-space kernel optimized using an iterative optimization algorithm and used in convolutional processing (second AI engine) is selected to improve the fidelity of the reconstruction and is determined by the undersampling pattern (unique sampling pattern) from the multi-coil undersampled k-space data (sparse-sampled MR scan acquisition sequence)).
Regarding claim 35, Mckinnon teaches further comprising a second AI engine configured to select each unique sampling pattern (Mckinnon col. 10, l. 39 – col. 11, l. 21, a k-space kernel optimized using an iterative optimization algorithm and used in convolutional processing (second AI engine) is selected to improve the fidelity of the reconstruction and is determined by the undersampling pattern (unique sampling pattern) from the multi-coil undersampled k-space data (sparse-sampled MR scan acquisition sequence)).
Regarding claim 36, Mckinnon teaches wherein the second AI engine was configured using a supervised learning algorithm (Mckinnon col. 10, ll. 23–30, 55–58, k-space convolution kernel is optimized (learning) based on known calibration data, thus supervised according to the calibration data).
Claims 10, 38, and 25 are rejected under 35 U.S.C. 103 as being unpatentable over Mckinnon in view of Zbontar in view of Principe, as set forth above, and further in view of Shaw et al., “Real-time three-dimensional MRI for the assessment of dynamic carpal instability,” April 7, 2019, PLoS ONE 14(9): e0222704, https://doi.org/10.1371/journal. pone.0222704 (herein “Shaw”).
Regarding claims 10 and 38, with claim 10 as exemplary, Mckinnon teaches the plurality of incomplete MR K-space scans (Mckinnon col. 7, ll. 41–63, MRI scan resulting in multi-coil undersampled k-space data) but does not explicitly teach where Shaw teaches are acquired using a static MR acquisition technique (Shaw page 3, Study subjects and positioning on scanner bed section, human subject was positioned in the MRI system and directed to hold their wrist motionless for 7 minutes for a static MRI acquisition).
Therefore, taking the teachings of Mckinnon as modified by Zbontar and Shaw together as a whole, it would have been obvious to a PHOSITA before the effective filing date of the claimed invention to have modified the MRI acquisition disclosed in Mckinnon to be static as disclosed in Shaw at least because doing so would provide a less sensitive acquisition method that can be performed with a higher signal-to-noise ratio and have higher spatial resolution. See Shaw Abstract.
Regarding claim 25, while Mckinnon teaches wherein each fully-sampled MR K-space data matrix is the result of an MR scan of an object (Mckinnon col. 25, ll. 18–38, combination of channel specific fully-sampled k-space data including processing of multiplying a coil image by the conjugate of its respective sensitivity map obtained from a fully sampled calibration region, where col. 10, ll. 11-12 and 23–30, col. 9, ll. 53-55, teach the calibration region begin part of the raw MRI data (result of scan) of a region of the subject (an object)), Mckinnon does not, where Shaw does teach an object obtained at the same position (Shaw page 3, Study subjects and positioning on scanner bed section, human subject was positioned in the MRI system and directed to hold their wrist motionless (same position) for 7 minutes for a static MRI acquisition).
Therefore, taking the teachings of Mckinnon as modified by Zbontar and Shaw together as a whole, it would have been obvious to a PHOSITA before the effective filing date of the claimed invention to have modified the MRI acquisition disclosed in Mckinnon to be in the same position as disclosed in Shaw at least because doing so would provide a less sensitive acquisition method that can be performed with a higher signal-to-noise ratio and have higher spatial resolution. See Shaw Abstract.
Claim 15 is rejected under 35 U.S.C. 103 as being unpatentable over Mckinnon in view of Zbontar in view of Principe, as set forth above, and further in view of Xue et al., “1D Probabilistic Undersampling Pattern Optimization for MR Image Reconstruction,” arXiv:2003.03797v3 [eess.IV] 9 Jan 2022 (herein “Xue”).
Regarding claim 15, Mckinnon does not explicitly teach, where Xue teaches wherein the probability distribution function is one of: a Gaussian probability distribution function and a Poisson probability distribution function (Xue page 1, right column, existing undersampling strategies using a Gaussian or Poisson undersampling strategy).
Therefore, taking the teachings of Mckinnon as modified by Zbontar and Xue together as a whole, it would have been obvious to a PHOSITA before the effective filing date of the claimed invention to have modified the MRI acquisition sampling pattern disclosed in Mckinnon to be Gaussian or Poisson as disclosed in Xue at least because doing so is conventional and used by many existing undersampling strategies having predictable results. Therefore such a modification would simply be simple substitution of one known element for another to obtain predictable results. See MPEP §2143(I)(B).
Claim 27 is rejected under 35 U.S.C. 103 as being unpatentable over Mckinnon in view of Zbontar in view of Principe, as set forth above, and further in view of Chen et al., US Patent Application Publication No. US 2021/0272297 A1 (herein “Chen”).
Regarding claim 27, while Mckinnon teaches wherein the loss function is a mean square function (Mckinnon col. 25, ll. 58–66, loss function as mean square error), Mckinnon does not teach where Chen teaches that it is a root mean square function (Chen ¶30, loss function for predicting MRI data being a root mean square) .
Therefore, taking the teachings of Mckinnon as modified by Zbontar and Chen together as a whole, it would have been obvious to a PHOSITA before the effective filing date of the claimed invention to have modified the MRI data prediction loss function disclosed in Mckinnon to be a root mean square loss function as disclosed in Chen at least because doing so would be obvious to try as there are only a limited number of commonly used loss functions in machine learning training, each with predictable results and a reasonable expectation of success. See MPEP §2143(I)(E).
Claim 37 is rejected under 35 U.S.C. 103 as being unpatentable over Mckinnon in view of Zbontar in view of Principe, as set forth above, and further in view of Guo et al., US Patent Application Publication No. US 2026/0051099 A1 (herein “Guo”).
Regarding claim 37, Mckinnon teaches the first and second AI engines (Mckinnon col. 10, ll. 23–30, 55–58, k-space convolution kernel (second AI engine) and col. 22, ll. 52–59, deep learning regularizer neural network (first AI engine)), however, Mckinnon does not where Guo teaches configured as a generative adversarial network (Guo ¶5, deep learning methods for medical image analysis including generative adversarial networks as well as convolutional neural networks).
Therefore, taking the teachings of Mckinnon as modified by Zbontar and Guo together as a whole, it would have been obvious to a PHOSITA before the effective filing date of the claimed invention to have modified the deep learning models as CNNs disclosed in Mckinnon to be a GAN as disclosed in Chen at least because doing so would be obvious to try as Guo in ¶5 provides a closed set of neural network structures used in image processing, each with predictable results and a reasonable expectation of success. See MPEP §2143(I)(E).
Claim 39 is rejected under 35 U.S.C. 103 as being unpatentable over Mckinnon in view of Zbontar in view of Principe, as set forth above, and further in view of Frahm et al., US Patent Application Publication No. US 2023/0280431 A1 (herein “Frahm”).
Regarding claim 39, Mckinnon teaches incomplete MR K-space data (Mckinnon col. 8, ll. 45–46, multi-coil undersampled k-space data from the MRI apparatus), but does not teach the remainder of the limitations of claim 37.
Zbontar teaches matrices comprises: two incomplete MR K-space data matrices for T1-weighted MR imaging; two incomplete MR K-space data matrices for T2-weighted MR imaging (Zbontar page 9, section 4.6, k-space data in the disclosed training dataset including 10,000 MRI DICOM studies (thus at least two of each) including T1 and T2 pulse sequences of 2D (matrices) image volumes (data of weighted MR imaging));
Frahm teaches and three incomplete MR K-space data matrices for diffusion-weighted MR imaging (Frahm Abstract, ¶¶33, 38 and 71, plurality of data samples generated with diffusion-weighted echo sequence and encoded into k-space and are undersampled (incomplete), where fig. 6 where fig. 6 shows three images under DW label).
Therefore, taking the teachings of Mckinnon as modified by Zbontar together as a whole, it would have been obvious to a PHOSITA before the effective filing date of the claimed invention to have modified the data disclosed in Mckinnon to be T1 and T2 data as disclosed in Zbontar at least because doing so would allow for training reconstruction on a larger variety of machines and settings. See Zbontar page 6, DICOM images section.
Further, taking the teachings of Mckinnon as modified by Zbontar and Frahm together as a whole, it would have been obvious to a PHOSITA before the effective filing date of the claimed invention to have modified the data disclosed in Mckinnon to be diffusion weighted data as disclosed in Frahm at least because doing so would allow for avoiding ambiguities with directional differences and increase the signal to noise ratio of the images, thus improving image quality. See Frahm ¶¶ 75, 73, and 45.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Lustig et al., US Patent Application Publication No. US 2010/0026294 A1, directed towards providing a magnetic resonance imaging signal using an oscillating gradient that applies blips in at least a second dimension in a pseudo-random order.
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MICHELLE M KOETH whose telephone number is (571)272-5908. The examiner can normally be reached Monday-Thursday, 09:00-17:00, Friday 09:00-13:00, EDT/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, Vincent Rudolph can be reached at 571-272-8243. 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.
MICHELLE M. KOETH
Primary Examiner
Art Unit 2671
/MICHELLE M KOETH/Primary Examiner, Art Unit 2671
Prosecution Insights