DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 01/26/2026 has been entered.
Response to Amendment
The amendment filed 01/15/2026 has been entered. Claims 1, 3-14, 17, 18, and 20 remain pending in the application, with claims 2, 16, and 19 having been cancelled.
Response to Arguments
Applicant’s arguments, pg. 8-14 of the Remarks filed 01/15/2026, with respect to Kalarot and Ahmad in the rejection of claim 1 have been considered but are moot because the new ground of rejection does not rely on any combination of references applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
The Remarks at the end of pg. 11 through the top of pg. 12 discuss the combination of Sureshjani and Amirrajab. These arguments have been fully considered, and the Examiner has updated the rationale in the rejection of claim 1 below. Applicant argues that “the asserted motivation to combine Amirrajab with Sureshjani is not valid under the MPEP” because the rationale is a “generic statement of a desired outcome rather than a reasoned explanation for modifying the method of Sureshjani to arrive at the claimed invention”. Further, Applicant argues “The cited passage merely reports observed performance improvements and does not teach or suggest restructuring the upstream synthesis method of Sureshjani to include downstream task-model training and storage,” thus rendering the asserted motivation not valid under the MPEP. In response, the Examiner points to the abstract and end of pg. 8 of Sureshjani, wherein the use of the generated synthetic images in “multi-site, multi-vendor” applications/scenarios is described. The last paragraph on pg. 8 through the top of pg. 9 specifically describes using the synthetic data generated by the method of Sureshjani to train an algorithm for segmentation, similar to the relied upon task model taught by Amirrajab. As previously cited from pg. 11 of Amirrajab, synthetic image data can be used to train the disclosed segmentation model. Thus, using the two relied upon references, a person of ordinary skill in the art would have been motivated to integrate the claimed limitations based on Sureshjani’s suggestion to utilize their synthetic images to train a task model and Amirrajab’s task model trained using synthetic images.
See MPEP 2143.01: “A "motivation to combine may be found explicitly or implicitly in market forces; design incentives; the ‘interrelated teachings of multiple patents’; ‘any need or problem known in the field of endeavor at the time of invention and addressed by the patent’; and the background knowledge, creativity, and common sense of the person of ordinary skill." Zup v. Nash Mfg., 896 F.3d 1365, 1371, 127 USPQ2d 1423, 1427 (Fed. Cir. 2018) (quoting Plantronics, Inc. v. Aliph, Inc., 724 F.3d 1343, 1354 [107 USPQ2d 1706] (Fed. Cir. 2013) (citing Perfect Web Techs., Inc. v. InfoUSA, Inc., 587 F.3d 1324, 1328 [92 USPQ2d 1849] (Fed. Cir. 2009) (quoting KSR, 550 U.S. at 418-21)).” See further MPEP 2143(I)(G): “The courts have made clear that the teaching, suggestion, or motivation test is flexible and an explicit suggestion to combine the prior art is not necessary. The motivation to combine may be implicit and may be found in the knowledge of one of ordinary skill in the art, or, in some cases, from the nature of the problem to be solved. Id. at 1366, 80 USPQ2d at 1649. "[A]n implicit motivation to combine exists not only when a suggestion may be gleaned from the prior art as a whole, but when the ‘improvement’ is technology-independent and the combination of references results in a product or process that is more desirable, for example because it is stronger, cheaper, cleaner, faster, lighter, smaller, more durable, or more efficient…” In view of the foregoing, the motivation to combine the relevant teachings is supported in both references as an explicit suggestion, and is further an implicit motivation that would be found in the knowledge of one of ordinary skill in the art (described in the claim 1 rejection below). Therefore, the combination of Sureshjani in view of Amirrajab is maintained.
Applicant’s arguments, pg. 14-19 of the Remarks filed 01/15/2026, with respect to claim 14 have been fully considered and are persuasive. See Allowable Subject Matter below.
Claim Objections
Claim 1 is objected to because of the following informalities: “machine training a task model” should read “machine training the task model”. Appropriate correction is required.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1, 3-4, 7, 9-10, and 12 are rejected under 35 U.S.C. 103 as being unpatentable over Abbasi-Sureshjani et al. (Abbasi-Sureshjani, S., Amirrajab, S., Lorenz, C., Weese, J., Pluim, J., & Breeuwer, M., 4D semantic cardiac magnetic resonance image synthesis on XCAT anatomical model (2020) in Medical Imaging with Deep Learning, Proceedings of Machine Learning Research, pp. 6-18), hereinafter Sureshjani, in view of Engel et al. (Engel, J., Hoffman, M., & Roberts, A. (2017). Latent constraints: Learning to generate conditionally from unconditional generative models. arXiv preprint arXiv:1711.05772.), hereinafter Engel, and Amirrajab et al. (Amirrajab, S., Al Khalil, Y., Lorenz, C., Weese, J., Pluim, J., & Breeuwer, M., Label-informed cardiac magnetic resonance image synthesis through conditional generative adversarial networks (2022) in Computerized Medical Imaging and Graphics, 101, 102123), hereinafter Amirrajab.
Regarding claim 1, Sureshjani teaches a method for machine learning for a cardiac magnetic resonance imaging task, the method comprising:
generating a synthetic sample of cardiac magnetic resonance imaging (Sureshjani, 1st sentence on pg. 4: “synthesize realistic-looking cardiac MR images”), the synthetic sample output by a machine-learned model (Sureshjani, pg. 4, 2nd sentence under Methodology section: “conditional image synthesis network is trained on real image data”) in response to input of a latent representation to the machine-learned model (Sureshjani, pg. 8, 2nd sentence: “Our style encoder encodes the local semantic information of the input style image, in addition to global style information, to a latent vector”; pg. 4, 3rd to last sentence in paragraph ‘Conditional image synthesis’: “Then the generator’s task is to combine the encoded (global and local) style and the content”).
Sureshjani fails to explicitly teach 1) the latent representation generated by an encoder that is trained independently of the machine-learned model and independently of a task model for the cardiac magnetic resonance imaging task, wherein the encoder is not part of an encoder-decoder network trained jointly with the machine-learned model and 2) machine training a task model for the cardiac magnetic resonance imaging task using the synthetic sample as training data; and storing the task model as machine trained (Sureshjani, mentioned in future works, see last sentence on pg. 9: “The goal is to investigate the utility of the synthetic data in training deep learning algorithm for segmentation”).
However, Engel teaches a similar system comprising an encoder and GAN (Engel, see generator and VAE in FIG. 1 caption on pg. 2 and FIG. 12 on pg. 19 wherein z is input to the generator). Engel teaches a generative model wherein the latent representation generated by an encoder that is trained independently of the machine-learned model and independently of a task model (Engel, pre-training of the encoder means that it is trained independently of the trained generator and any subsequently implemented models, such as a task model taught in the combination with Amirrajab below; FIG. 1 caption on pg. 2: “(b) We begin by pretraining a standard VAE, with an emphasis on achieving good reconstructions”; see FIG. 1 attached below), wherein the encoder is not part of an encoder-decoder network trained jointly with the machine-learned model (Engel, VAE is pre-trained, and thus not trained jointly with the machine learned model).
Sureshjani discloses an encoder and generator, but does not disclose wherein the encoder is trained independently; however, Engel teaches a pre-trained encoder in a similar system. A person of ordinary skill in the art, before the effective filing date of the claimed invention, would have recognized that the pre-trained encoder, taught by Engel, could have been substituted for the encoder of Sureshjani because both serve the purpose of generating latent representations for input to an image generator model. Furthermore, a person of ordinary skill in the art would have been able to carry out the substitution. Therefore, it would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to substitute the pre-trained encoder in the above teachings of Engel for the encoder of Sureshjani in order to generate realistic images, having the same attributes as the original images (Engel, 3rd bullet point on pg. 2: “Because we start from a VAE that can reconstruct inputs well, we are able to apply identity-preserving transformations by making the minimal adjustment in latent space needed to satisfy the desired constraints. For example, when we adjust a person’s expression or hair, the result is still clearly identifiable as the same person”; FIG. 2 caption on pg. 3: “Using an actor to shift prior samples to satisfy the realism constraint, we achieve more realistic samples without sacrificing sharpness (Row 6). The samples are mapped to the closest point in latent space that both satisfies the realism constraint and has the same attributes as the original data”).
PNG
media_image1.png
568
719
media_image1.png
Greyscale
Further, Amirrajab teaches machine training a task model (Amirrajab, “DL model” in citation below) for the cardiac magnetic resonance imaging task (Amirrajab, “segmentation” in citation below) using the synthetic sample as training data; and storing the task model as machine trained (Amirrajab, pg. 7, 1st sentence in left column: “We employ synthesized images for training a DL model to segment left and right ventricular cavities and myocardium of the heart”; machine trained task model is the trained DL model). It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to combine the trained task model of Amirrajab with the method of Sureshjani in view of Engel in order to evaluate the generated synthetic images using a segmentation task model (Sureshjani, last para of pg. 8: “quantitative application-based evaluation of the synthetic images by deploying them in a heart segmentation task for multi-site, multi-vendor scenarios…utilize the synthetic images in different data augmentation strategies for the cardiac cavity segmentation task. The goal is to investigate the utility of the synthetic data in training deep learning algorithm for segmentation and evaluate that the data generated by this approach is clinically meaningful to replace the need for real data.”). Sureshjani suggests the use of a task model, while Amirrajab teaches a machine trained segmentation task model trained using synthetic samples (Amirrajab, pg. 11, 4th and 7th sentences under section ‘6.3’: “the addition of the synthetic data can substantially improve the model performance…aiding the training with multi-tissue synthetic images yields a better segmentation performance around the apex and base of the heart and more accurate segmentation of the right ventricular cavity and myocardium”).
Furthermore, the use of synthetic images to train a machine learned task model is obvious to try (see MPEP 2143(I)(E)). As described by both references, there is a recognized need in the art to increase the amount of diverse training images to train computer vision machine learning models (Sureshjani, abstract: “this approach can result in a realistic virtual population to address different challenges the medical image analysis research community is facing such as expensive data collection…synthesize 4D controllable CMR images with annotations and adaptable styles to be used in various supervised multi-site, multi-vendor applications in medical image analysis”; Amirrajab, abstract: “Synthesis of a large set of high-quality medical images with variability in anatomical representation and image appearance has the potential to provide solutions for tackling the scarcity of properly annotated data in medical image analysis research”). Since an increase in medical images is required, a predictable potential solution to this problem is to utilize synthetically generated images, as opposed to real medical images captured by a medical imaging device. This is a known potential solution with a reasonable expectation of success, further suggested by Sureshjani and Amirrajab (see citations in the last paragraph – Sureshjani suggests the utility of synthetic data in training deep learning algorithms to replace the need for real data, while Amirrajab demonstrates this utilization to improve a segmentation model). Therefore, a person having ordinary skill in the art would have found combining the trained task model of Amirrajab with the method of Sureshjani in view of Engel as obvious to try, from a finite number of identified, predictable solutions, in order to successfully increase the accessibility to a diverse set of medical images for training a task model.
Regarding claim 3 (dependent on claim 1), Sureshjani in view of Engel and Amirrajab teaches wherein generating the synthetic sample comprises generating where the latent representation is generated as a representation of style of a cardiac magnetic resonance image (Sureshjani teaches encoding the style of cardiac MRI, pg. 8, 2nd sentence: “Our style encoder encodes the local semantic information of the input style image, in addition to global style information, to a latent vector”; pg. 7, 4th sentence: “The input images of the encoder (representing the style) are depicted in the first column”, see CMRI images in Figure 4 on pg. 8; In the combination with Engel, Engel similarly teaches wherein the encoder preserves the features of the input image, 3rd bullet point on pg. 2: “we are able to apply identity preserving transformations by making the minimal adjustment in latent space needed to satisfy the desired constraints. For example, when we adjust a person’s expression or hair, the result is still clearly identifiable as the same person”).
Regarding claim 4 (dependent on claim 1), Sureshjani in view of Engel and Amirrajab teaches wherein generating the synthetic sample comprises generating by the machine-learned model comprising a generator having been trained as a generative adversarial network (Sureshjani, pg. 1, 2nd sentence of Abstract: “synthesizes CMR images via a data-driven Generative Adversarial Network”; see training of “SPADE GAN” in Figure 1 on pg. 5).
Regarding claim 7 (dependent on claim 1), Sureshjani in view of Engel and Amirrajab teaches wherein generating the synthetic sample comprises generating the synthetic sample and additional synthetic samples by variation of the latent representation input to the machine-learned model (Sureshjani, 5th sentence on pg. 7: “Two different synthetic images for each style are shown in the second and third columns, and the label maps (the inputs of the SPADE layers) are shown on the top left corner of the resulting synthetic images”, see outputs for the two different inputted styles in Figure 4 on pg. 8; see further, last sentence on pg. 8: “We use our proposed approach to generate a large virtual population with various anatomical and style variations”).
Regarding claim 9 (dependent on claim 1), Sureshjani in view of Engel and Amirrajab teaches wherein generating the synthetic sample comprises generating cardiac magnetic resonance images as a plurality of slices at different times (Sureshjani, Figure 3 on pg. 7: “4D synthetic images on XCAT labels for 12 time frames from end-diastolic to end-systolic phase”; each image is a slice; Figure 7 on pg. 13 also shows a plurality of slices, apex, mid, and base location slices, each over 12 time frames).
Regarding claim 10 (dependent on claim 1), Sureshjani in view of Engel and Amirrajab teaches wherein generating the synthetic sample comprises generating where the latent representation is generated by a machine-learned autoencoder having been trained with a loss based on comparison of an output image with a ground truth image and based on comparison of latent representations (Engel, see VAE introduced in claim 1 and regularized loss at the bottom of pg. 5).
Regarding claim 12 (dependent on claim 1), Sureshjani in view of Engel and Amirrajab teaches wherein machine training comprises machine training the task model for segmentation, ejection fraction computation, disease classification, plane classification, view classification, or landmark detection (Amirrajab, segmentation, pg. 7, 1st sentence in left column: “We employ synthesized images for training a DL model to segment left and right ventricular cavities and myocardium of the heart”).
Claim 5 is rejected under 35 U.S.C. 103 as being unpatentable over Sureshjani in view of Engel, Amirrajab, Diller et al. (Diller, G. P., Vahle, J., Radke, R., Vidal, M. L. B., Fischer, A. J. et al., Utility of deep learning networks for the generation of artificial cardiac magnetic resonance images in congenital heart disease (2020) in BMC medical imaging, 20, 1-8), hereinafter Diller, and Lyu et al. (Lyu, Q., Shan, H., Xie, Y., Kwan, A. C., Otaki, Y., Kuronuma, K. et al., Cine cardiac MRI motion artifact reduction using a recurrent neural network (2021) in IEEE transactions on medical imaging, 40(8), 2170-2181), hereinafter Lyu.
Regarding claim 5 (dependent on claim 4), Sureshjani in view of Engel and Amirrajab teaches wherein generating the synthetic sample comprises generating where the generative adversarial network was a conditional generative adversarial network (Sureshjani, SPADEGAN, pg. 4, Methodology section under heading ‘Conditional image synthesis’: “conditional image synthesis network”, see SPADEGAN details in subsequent paragraph), but fails to teach where the generative adversarial network was a recurrent progressive conditional generative adversarial network.
However, Diller teaches a progressive generative adversarial network (Diller, pg. 3, left column, “Progressive GAN”). It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to combine the progressive GAN components of Diller (progressively increasing image size, see further “Progressive GAN” section on pg. 3 of Diller) with the method of Sureshjani in view of Engel and Amirrajab in order to generate high resolution synthetic images (Diller, pg. 7, 1st paragraph in left column: “start with a low-resolution GAN and increasing image size step by step during training…thereby supporting the generator and stabilizing the model”).
Further, Lyu teaches a recurrent generative adversarial network (Lyu, pg. 4, 1st sentence: “We propose a recurrent generative adversarial network model for cardiac MRI motion artifact reduction”; see Figure 2 on pg. 18). It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to combine the recurrent GAN architecture of Lyu (incorporating recurrent generator components, see Figure 2 of Lyu) with the method of Sureshjani in view of Engel and Amirrajab in order to capture the relationship between dynamic cardiovascular images (Lyu, 3rd to last sentence of ‘Abstract’ on pg. 2: “We demonstrate a decent generalizability of the proposed method thanks to the novel architecture of our deep network that captures the essential relationship of cardiovascular dynamics”). Additionally, Sureshjani teaches the generation of time series data (See Figure 6 below from pg. 12 of Sureshjani). It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to utilize the recurrent GAN architecture of Lyu to produce time series CMRI images (Lyu, abstract on pg. 2: “This model utilizes bi-directional convolutional long short-term memory (ConvLSTM) and multi-scale convolutions to improve the performance of the proposed network, in which bi-directional ConvLSTMs handle long-range temporal features while multiscale convolutions gather both local and global features”).
PNG
media_image2.png
458
787
media_image2.png
Greyscale
Claim 6 is rejected under 35 U.S.C. 103 as being unpatentable over Sureshjani in view of Engel, Amirrajab and Lyu.
Regarding claim 6 (dependent on claim 1), Sureshjani in view of Engel and Amirrajab teaches wherein generating the synthetic sample comprises generating by the machine-learned model comprising an up sampling deep neural network (Sureshjani, pg. 4, 2nd sentence in paragraph ‘Conditional image synthesis’: “a series of the residual blocks with SPADE normalization, followed by nearest neighbor up-sampling layers”) and a styling deep neural network (Sureshjani, performed by the generator synthesizing images using style information, pg. 4, 3rd to last sentence in paragraph ‘Conditional image synthesis’: “Then the generator’s task is to combine the encoded (global and local) style and the content coming from the semantic segmentation mask to synthesize an image. This setup is useful in controlling the style of synthetic images and the reconstruction of the surrounding organs of the heart”), but fails to teach a plurality of up sampling and styling deep neural networks, and a plurality of long-short term memories.
Lyu teaches a plurality of neural networks (Lyu, encoder and decoder blocks in a model that each generate an image in a time series, Fig. 3 on pg. 19), and a plurality of long-shot term memories (Lyu, abstract on pg. 2: “This model utilizes bi-directional convolutional long short-term memory (ConvLSTM)”; see plurality of LSTM blocks in Fig. 3 on pg. 19). It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to combine the plurality of neural network blocks and long-short term memories of Lyu with the up sampling and styling deep neural networks and model architecture of Sureshjani (create a recurrent model architecture similar to Lyu), respectively, in the method of Sureshjani in view of Engel and Amirrajab in order to improve image quality (Lyu, 3rd sentence on pg. 9: “Each frame in the sequence was fed into the encoder, and the resultant features were concatenated with those generated from the ConvLSTM branches in each frame state”; temporal information feeds into each encoder, improving image quality of each slice generated, pg. 11, 3rd sentence under ‘Discussions’: “incorporating ConvLSTM branches and involving temporal information in the networks leads to the image quality better than that only based on spatial information”). Additionally, Sureshjani teaches the generation of time series data (See Figure 6 of Sureshjani in the claim 5 rejection). It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to utilize the recurrent GAN architecture of Lyu to produce time series CMRI images (Lyu, abstract on pg. 2: “This model utilizes bi-directional convolutional long short-term memory (ConvLSTM) and multi-scale convolutions to improve the performance of the proposed network, in which bi-directional ConvLSTMs handle long-range temporal features while multiscale convolutions gather both local and global features”).
Claim 8 is rejected under 35 U.S.C. 103 as being unpatentable over Sureshjani in view of Engel, Amirrajab, and Peña et al. (Peña, H., Gómez, S., Romo-Bucheli, D., & Martinez, F., Cardiac disease representation conditioned by spatio-temporal priors in cine-MRI sequences using generative embedding vectors (2021) in 2021 43rd Annual International Conference of the IEEE Engineering in Medicine & Biology Society, pp. 5570-73), hereinafter Peña.
Regarding claim 8 (dependent on claim 1), Sureshjani in view of Engel and Amirrajab teaches wherein generating the synthetic sample comprises generating with the input to the machine-learned model of the latent representation (Sureshjani, pg. 8, 2nd sentence: “Our style encoder encodes the local semantic information of the input style image, in addition to global style information, to a latent vector”; pg. 4, 3rd to last sentence in paragraph ‘Conditional image synthesis’: “Then the generator’s task is to combine the encoded (global and local) style and the content”), but fails to teach input of values for one or more parameters, the one or more parameters comprising a pathology, base and apex indices, an electrocardiogram, an ejection fraction, or a number of slices.
However, Peña teaches input to a machine-learned model of values for one or more parameters comprising a pathology (Peña, pg. 5571, 4th sentence under header ‘Deep conditioned and generative representation’: “In the MCIGAN, the generator is fed with a random vector in the latent space, the associated heart pathology and vector information related with prior structural or motion information”; see Fig. 2 on pg. 5571). It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to combine the pathology input parameter of Peña with the method of Sureshjani in view of Engel and Amirrajab in order to produce a GAN model that can differentiate between cardiac pathologies (Peña, pg. 5570, 1st sentence in last paragraph: “This work presents a strategy based on a conditional GAN architecture, conditioned by cardiac prior information, to recover embedding patterns and to differentiate among five different cardiac pathologies”).
Claim 11 is rejected under 35 U.S.C. 103 as being unpatentable over Sureshjani in view of Engel, Amirrajab, and Beetz et al. (Beetz, M., Banerjee, A., & Grau, V., Multi-domain variational autoencoders for combined modeling of MRI-based biventricular anatomy and ECG-based cardiac electrophysiology (2022) in Frontiers in physiology, 13, 886723), hereinafter Beetz.
Regarding claim 11 (dependent on claim 1), Sureshjani in view of Engel and Amirrajab teaches wherein generating the synthetic sample comprises generating with the input comprising the latent representation (Sureshjani, pg. 4, 3rd to last sentence in paragraph ‘Conditional image synthesis’: “Then the generator’s task is to combine the encoded (global and local) style and the content”) and a cardiac magnetic resonance image (Sureshjani, pg. 4, 2nd sentence in paragraph ‘The real dataset’: “This dataset consists of Cine MR images”; see ACDC cardiac data input in Figure 1 on pg. 5), but fails to teach wherein the input comprises an electrocardiogram.
However, Beetz teaches generating a cardiac model (Beetz, abstract) with an electrocardiogram (Beetz, pg. 2, 1st bullet point in the right column: “We propose a novel multi-domain variational autoencoder capable of modeling combined cardiac anatomy and ECG data”; pg. 4, 1st sentence in section ‘2.3’: “architecture with three branches that share a common latent space for inter-modal information sharing”; see ECG input into model in Figure 1c on pg. 3). It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to combine the electrocardiogram input parameter of Beetz with the method of Sureshjani in view of Engel and Amirrajab in order to improve the identification of cardiovascular disease (Beetz, 2nd to last sentence in ‘Conclusion’ on pg. 13: “we have observed that combined anatomy and ECG representations improve the identification of cardiovascular disease compared to single-domain approaches”).
Claim 13 is rejected under 35 U.S.C. 103 as being unpatentable over Sureshjani in view of Engel, Amirrajab and Shayani et al. (U.S. Patent No. 2023/0326158 A1), hereinafter Shayani.
Regarding claim 13 (dependent on claim 1), Sureshjani in view of Engel and Amirrajab teaches wherein machine training further comprises machine training the task model with input of the synthetic sample (Amirrajab, pg. 7, 1st sentence in left column: “We employ synthesized images for training a DL model”), but fails to teach input of the latent representation (Sureshjani teaches wherein the latent vector is inputted to the generator, but not input to a task model).
However, Shayani teaches machine training a task model with input of the latent representation (Shayani, para 80: “style-generation engine 126 can train machine learning model 228 to learn a more complex style code distribution 230, given the set of style codes”; style codes are latent space, para 82: “The embedding space associated with the style codes in the pairs could correspond to the latent space of the style codes”). It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to combine the method of training a model with input of a latent representation of Shayani with the method of Sureshjani in view of Engel and Amirrajab in order to learn a distribution of style information (Shayani, para 80: “could train one or more components of a…and/or another type of generative model to learn a latent manifold corresponding to style code distribution 230 within the latent space occupied by style codes generated by style network 214”) and/or train a model to disentangle style and content information (Shayani, para 117: “with the disclosed techniques, attributes pertaining to style in 3D shapes can be disentangled from attributes pertaining to content”).
Allowable Subject Matter
Claims 14, 17, 18, and 20 are allowed. The statement of reasons for the indication of allowable subject matter described in the Final Rejection of 12/05/2025 applies to amended claim 18.
Regarding claim 14, while Sureshjani in view of Amirrajab, Peña, and Zhang teach the input of latent variables for number of slices, slice position relative to anatomy, pathology, and functional measurement to the machine-learned model, the relied upon prior art fails to teach in reasonable combination wherein the latent representation generated by the encoder (for example, the encoder in the rejection of claim 1 taught by Sureshjani in view of Engel and Amirrajab) comprises all types of the listed values. Therefore, the prior art fails to teach as a whole generating synthetic samples of cardiac magnetic resonance imaging, the synthetic samples output by a machine-learned model in response to input of a latent representation comprising values for a number of slices, pathology, functional measurement, and slice position relative to anatomy, the latent representation generated by an encoder that is trained independently of the machine-learned model and independently of a task model for the cardiac magnetic resonance imaging task; machine training the task model for the cardiac magnetic resonance imaging task using the synthetic samples as training data; and storing the task model as machine trained.
In view of the foregoing, the prior art references alone or in reasonable combination are insufficient to teach the invention as a whole, as claimed in claim 14. Due to its dependence on claim 14, claim 17 is similarly objected to.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
Meshry et al. (Meshry, M., Ren, Y., Davis, L. S., & Shrivastava, A. (2021). Step: Style-based encoder pre-training for multi-modal image synthesis. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (pp. 3712-3721).) teaches an image synthesis model wherein the encoder is trained independently of the generator model and not part of an encoder-decoder network trained jointly with the generator model; see the attachment below, specifically the Stage 1-3 bullet points.
PNG
media_image3.png
762
510
media_image3.png
Greyscale
Shang et al. (Shang, W., & Sohn, K. (2019, January). Attentive conditional channel-recurrent autoencoding for attribute-conditioned face synthesis. In 2019 IEEE Winter Conference on Applications of Computer Vision (WACV) (pp. 1533-1542). IEEE.) teaches a recurrent progressive conditional GAN (abstract: “Building on top of a conditional version of VAE-GAN, we augment the pathways connecting the latent space with channel-recurrent architecture…Lastly, we incorporate the progressive-growth training scheme to the inference, generation and discriminator networks of our models to facilitate higher resolution outputs.”; see recurrent components in generation network in FIG. 2C on pg. 1534; right column on pg. 1534: “propose our final model, the attentive conditional channel-recurrent VAE-GAN(acVAE-GAN)”).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to EMMA E DRYDEN whose telephone number is (571)272-1179. The examiner can normally be reached M-F 9-5 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, ANDREW BEE can be reached on (571) 270-5183. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of 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.
/EMMA E DRYDEN/Examiner, Art Unit 2677 /ANDREW W BEE/Supervisory Patent Examiner, Art Unit 2677