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and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114.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 06/24/2026 has been entered.
Response to Amendment
Amendments dated 06/24/2026 are accepted.
Claims 1-20 are pending.
Claims 1, 3-6, 8-11,13-14,16, and 19- 20 are amended.
Regarding rejection under U.S.C. 35 §101
Applicant notes rejection of claims 1-20, dated 04/24/2026 was withdrawn as indicated in Examiner’s Advisory Action response dated 05/07/2026, based on applicant arguments in response after final action dated 04/24/2026, specifically related to recent guidance based on Ex Parte Desjardins: rejection of claims 1-20 under U.S.C. 35 §101 is withdrawn.
Regarding rejection under U.S.C. 35 §103 over prior art
Applicant arguments have been fully considered but are not persuasive. Specifically, arguments are directed to claims as currently amended, which present matter that has not been previously considered, and thus requires further search and evaluation.
Specifically, Applicant argues regarding rejection of Claim 1 over prior art by BUCHHOLZ that cannot be relied upon to teach claim as presently amended, which now includes language of “updating one or more parameters of a second trained machine learning model based on the conditioning information to generate an updated model that represents the first physical object within the spectral domain” (Remarks, P7, last paragraph), i.e., that BUCHHOLZ does not teach a method of reconstructing an image of a physical object. Examiner respectfully disagrees, pointing to BUCHHOLZ reconstruction of samples, where samples are physical objects represented in image data related to CT scanning processes. See BUCCHOLZ Pg 2, ss 2.4 Tomographic image Reconstruction” and Fig. 6 reconstruction of an image of human heads, which are “representations of physical objects” However, as noted, the amended language in claims as currently presented, has not been previously considered, necessitating additional search and evaluation. Further search revealed references such that claims as currently amended do not differentiate over prior art, with more explicit reference to reconstruction of physical objects. New grounds of rejection, as necessitated by amendment, is presented in detail below.
Claim Rejections - 35 USC § 103
The following is a quotation of pre-AIA 35 U.S.C. 103(a) which forms the basis for all obviousness rejections set forth in this Office action:
(a) A patent may not be obtained though the invention is not identically disclosed or described as set forth in section 102, if the differences between the subject matter sought to be patented and the prior art are such that the subject matter as a whole would have been obvious at the time the invention was made to a person having ordinary skill in the art to which said subject matter pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-9 and 11-20 are rejected under pre-AIA 35 U.S.C. 103(a) as being unpatentable over BUCHHOLZ (Buchholz, et al., "Fourier Image Transformer " arXiv:2104.02555v2 [cs.CV] 3 May 2021), in view of PEREZ (Perez, et. al., “FiLM: Visual Reasoning with a General Conditioning Layer”, arXiv:1709.07871v2, 18 Dec 2017) , and further in view of MEYERS-NORMAND (US 20220181007 A1)
With respect to independent Claims 1, 11, 20, BUCHHOLZ teaches:
(Claim 1) A computer-implemented method
(Claim 11): One or more non-transitory computer readable media including instructions that, when executed by one or more processors, cause the one or more processors
(Clam 20) A system comprising: one or more memories storing instructions; and
one or more processors coupled to the one or more memories that, when executing the instructions, perform the steps of:
for reconstructing representations of physical objects in a spectral domain, (BUCHHOLZ is in same technical field, Abstract: “a sequential image representation… Fourier Domain Encodings (FDEs), an autoregressive image completion task is equivalent to predicting a higher resolution output given a low-resolution input”; and Section 3.4: “Fourier Image Transformer setup for tomographic reconstruction (“FIT: TRec”); Examiner interprets “spectral domain” as analogous to the method of using frequency domain with Fourier methods, as would be understood by one of ordinary skill; BUCHHOLZ teaches computer-implemented method for image reconstruction, Fig. 3 with caption, “encoder-decoder based Fourier Image Transformer setup for tomographic reconstruction…2D computed tomography, 1D projections of an imaged sample…are back-transformed into a 2D image.”; Examiner interprets “physical object” as analogous to “imaged sample”; BUCHHOLZ teaches non-transitory computer readable media and processors to execute instructions, as would be understood by one of ordinary skill, Abstract: “computed tomography image reconstruction”, 2.3. “Transformers in Computer Vision”, or 2.4. “Tomographic Image Reconstruction”))
using a first trained machine learning model that maps a first set of data points associated with both a first physical object and the spectral domain to conditioning information (BUCHHOLZ teaches trained machine learning model used for mapping, Fig. 2, with Pg1, 1: “encoder-decoder based Fourier Image Transformer (“FIT: TRec”) can be trained on a set of Fourier measurements and then used to query arbitrary Fourier coefficients, which we use to improve sparse-view computed tomography (CT) image restoration mapping”; BUCHHOLZ teaches use of conditioning information and transformation to frequency domain, Pg2,2.3: “first n pixels of the flattened input image are used to condition a generative transformer setup that then predicts the remaining image in an auto-regressive manner”, and Pg.3, Section 3.1, FIG.2 and caption: “Low-resolution input images are first transformed into Fourier space and then unrolled into an FDE sequence…fed to a FIT, that, conditioned on this input, extends the FDE sequence to represent a higher resolution image…FIT is conditioned on the first 39 entries of the FDE”; Examiner interprets “data points associated with a first physical object” using BRI to be analogous to reference Pg2,§2.4 “Tomographic image reconstruction”, “2D sample section is acquired by rotating a 1D detector array around
the sample, acquiring a series of density measurements” to mean a representation of a physical object based on its density. )
updating one or more parameters of a second trained machine learning model (BUCCHOLZ teaches iterative model development, generally, Pg.1, “Section 3 we introduce our novel Fourier Domain Encoding (FDE) and training strategies for auto-regressive and encoder-decoder transformer models”; and Pg. 6, Table 1, discloses details of iterative, progressing model development, and §5: “Discussion…idea of Fourier Domain Encodings (FDEs),a novel sequential image encoding”; Examiner interprets “updating one or more parameters” to mean generally, the iterative process disclosed by BUCCHOLZ, asserting that one of ordinary skill would understand updating parameters to use the same core processes as training, in an iterative processes of machine learning.)
generate an updated model that represents the first physical object within the spectral domain; (BUCCHOLZ teaches generative model in the spectral domain, as above, P4, SS3.3: “final prediction images
x
^
are generated by computing the inverse Fourier transform on predictions
C
^
, which are rearranged (rolled) into
X
^
h
and completed to a full predicted Fourier spectrum
X
^
”, and as above, Section 3.4: “Fourier Image Transformer setup for tomographic reconstruction (“FIT: TRec”); Examiner interprets “spectral domain” and “represents the physical object” as discussed as analogous to reference.)
constructing an image associated with the first physical object (BUCHHOLZ teaches reconstruction for each dataset, as above, Fig. 6, Pg6, 4.6; and predicted reconstruction, Pg.4 Section 3.4: “..such that the predicted reconstruction
PNG
media_image1.png
16
14
media_image1.png
Greyscale
of
PNG
media_image2.png
11
13
media_image2.png
Greyscale
can be computed by
PNG
media_image3.png
24
146
media_image3.png
Greyscale
, where roll arranges the 1D sequence back into a discrete 2D Fourier spectrum…“FIT: TRec””; Examiner interprets “image associated with the first physical object” as above, pointing to Fig6, wherein images of physical objects (human heads) are shown as reconstructed by the method)
BUCCHOLZ does not explicitly teach:
generating conditioning information using a first trained machine learning model
updating one or more parameters of a second trained machine learning model based on the conditioning information
generating a second set of data points associated with both the first item and the spectral domain via the model
reconstruction associated with the first physical object based on the second set of data points.
PEREZ teaches:
generating conditioning information using a first trained machine learning model (PEREZ is in same technical field, Abstract: “general-purpose conditioning method for neural networks called FiLM: Feature-wise Linear Modulation… answering image-related questions which require a multi-step, high-level process”; PEREZ teaches explicitly several methods for generating conditioning information using trained model, Pg.2,§2 “Method”, and Pg.4, §4 “Experiments”)
It would have been obvious to one of ordinary skill in the art before effective filing date of the claimed invention to modify BUCHHOLZ to include a first model that generates condition information, as taught by PEREZ because it would result in improved mapping by adding responsive flexibility based on changes in the conditioning information. Examiner notes BUCCHOLZ does teach the idea of conditioning model (BUCCHOLZ Pg.2,SS2.3, as above) but does not explicitly teach how the conditioning information is achieved, as does PEREZ. One of ordinary skill would see the obvious connection and advantage of using the ideas disclosed by PEREZ with the suggestion of BUCCHOLZ to improve the method and system for reconstruction. One of ordinary skill would understand that generation of conditioning data would ultimately allow for more effective control of the generative process for image reconstruction, particularly for inputs with low resolution or sparse data sets and would see that value of using a pre-trained model to provide conditioning information to accelerate the training process of the second model or make it more data-efficient.
BUCHHOLZ, as modified by PEREZ, as taught above does not teach:
updating one or more parameters of a second trained machine learning model based on the conditioning information
generating a second set of data points associated with both the first item and the spectral domain via the model
reconstruction associated with the first physical object based on the second set of data points.
MEYERS-NORMAND teaches:
updating one or more parameters of a second trained machine learning model based on the conditioning information; (MEYERS-NORMAND is in related technical teaching a machine-learning based image analysis and reconstruction method/system, Abstract: “device can train a first machine learning model (“model”) based on a first group of the retinal images”; Examiner interprets reference to “retinal images” as a representation of a physical object; and [0002]: “relates to deep learning…computing systems and processes for utilizing and training a computing system to predict geographic atrophy progression based on retinal images.”, and using spectral domain, [0047]: “retinal image includes…Spectral domain-optical coherence tomography (SD-OCT) retinal images”; MEYERS-NORMAND teaches using conditioning-based training method, FIG. 2 and [0037]: “machine learning system 200 includes a data conditioning module 212” ; Examiner notes BUCCHOLZ teaches use of conditioning information but does not explicitly disclose incorporation into second model, as does MEYERS-NORMAND.)
generating a second set of data points associated with both the first item and the spectral domain via the model (MEYERS-NORMAND teaches iteratively trained models, Abstract: “raining the first model and the second model, the device can train a third model to predict”, and FIG.4, and FIG.8, with [0009]: “instructions for training a first machine learning model based on a first group of the retinal images and patient data corresponding to the first group, wherein the first group includes a first type of retinal image…instructions for training a second machine learning model…instructions for generating, using the trained second machine learning model, a second prediction of geographic atrophy progression based on a second subset of the third group of the retinal images and patient data corresponding to the second subset. After training the first machine learning model and the second machine learning model, the one or more programs can also include instructions for training a third machine learning model…based on the first prediction, the second prediction, the first subset, the second subset and patient data corresponding to the first subset and the second subset.”))
reconstruction associated with the first physical object based on the second set of data points. (As above, MEYERS-NORMAND teaches generative model with prediction outputs based on first item and second model, [0009], and [0008]: “generating, using the trained first machine learning model, a first prediction of geographic atrophy progression based on a first subset of a third group of the retinal images and patient data corresponding to the first subset…generating, using the trained second machine learning model, a second prediction of geographic atrophy progression based on a second subset”; Examiner interprets “reconstruction associated with first physical object” as above, analogous to reference, for example Fig. 5 depicts a reconstruction after processing of a retinal surface, i.e., “physical object”.)
It would have been obvious to one of ordinary skill in the art before effective filing date of the claimed invention to modify BUCHHOLZ, as modified by PEREZ and taught above, to include updating one or more parameters of a second trained machine learning model based on the conditioning information, generating a second set of data points associated with both the first item and the spectral domain via the model, and reconstruction associated with the first physical object based on the second set of data points, as taught by MEYERS-NORMAND, because these techniques take advantage of unique properties contained in frequency domain data in a wide range of imaging application fields. One of ordinary skill would understand the advantage of these explicit steps in a robust machine-learning methods as an obvious way to leverage the method taught by BUCCHOLZ as modified by PEREZ to improve removal of undesired artifacts, including blur and noise, and to improving the ability to extract meaningful data even from low signal-to-noise or sparsely constructed input image data. acquisition. One of ordinary skill would be motivated to combine the explicit disclosure of machine-learning methods using multiple iterative steps and conditioning information incorporation into a trained model, as taught by MEYERS-NORMAND with the method of image reconstruction using the related FOURIER techniques as taught by BUCHHOLZ to yield improved reconstructed image, even for sparse input data by allowing the generative model to follow intentional structural constraints.
With respect to Claims 2 and 12, BUCHHOLZ, as modified by PEREZ and MEYERS-NORMAND, as taught above, teaches the limitations of Claims 1 and 11.
BUCHHOLZ further teaches:
wherein mapping the first set of data points to the conditioning information comprises performing one or more positional encoding operations on the first set of data points. (BUCHHOLZ teaches mapping and conditioning in an encoding process, Fig. 2 and Pg1, 1 and Pg2,2.3)
With respect to Claims 3 and 13, BUCHHOLZ, as modified by PEREZ and MEYERS-NORMAND as taught above, teaches the limitations of Claims 1 and 11.
BUCHHOLZ, further teaches:
wherein updating the one or more parameters of second trained machine learning model comprises modifying one or more values of one or more parameters associated with the second trained machine learning model based on the conditioning information. (BUCHHOLZ teaches iterative process, see Pg1-2,2.1: “an encoder-decoder structure, where the encoder maps frequency Fourier coefficients, an input encoding
PNG
media_image4.png
33
153
media_image4.png
Greyscale
into a continuous latent space
PNG
media_image5.png
33
157
media_image5.png
Greyscale
, with N corresponding to the number of input tokens and F representing the feature dimensionality per token…latent space embedding z is then given to the decoder, which generates an M long output sequence
PNG
media_image6.png
33
127
media_image6.png
Greyscale
iteratively, element by element…auto-regressive decoding scheme means that the decoder generated the i-th output token while not only observing z, but also all i−1 output tokens generated previously.”; and, as above, BUCHHOLZ teaches using conditioning, Pg3, Fig.2; and teaches model construction details, Pg3.3: “Methods”.)
BUCCHOLZ, as modified by PEREZ and MEYERS-NORMAND as taught above, does not teach:
updating the second trained machine learning model comprises modifying one or more values of one or more parameters associated with the second trained machine learning model based on the conditioning information.
MEYERS-NORMAND further teaches:
updating the second trained machine learning model comprises modifying one or more values of one or more parameters associated with the second trained machine learning model based on the conditioning information. (As above, MEYERS-NORMAND teaches repetitive use of conditioning module in iterative model progression, FIG.4, and [0008]: “generating, using the trained first machine learning model, a first prediction of geographic atrophy progression based on a first subset of a third group of the retinal images and patient data corresponding to the first subset…generating, using the trained second machine learning model, a second prediction of geographic atrophy progression based on a second subset”)
It would have been obvious to one of ordinary skill in the art before effective filing date of the claimed invention to modify BUCCHOLZ, as modified by PEREZ and MEYERS-NORMAND, as taught above, to include the technique of updating the second trained machine learning model comprises modifying one or more values of one or more parameters associated with the second trained machine learning model based on the conditioning information, as further disclosed by MEYER-NORMAND because these steps add iterative modeling details that would result in a more accurate model for improving the image reconstruction method/system of BUCCHOLZ as modified above. One of ordinary skill would be familiar with the advantages of including adaptability steps for developing iterative models where each model is improved from the previous model. One of ordinary skill would see an obvious connection between the explicit training steps disclosed by MEYERS-NORMAN which avoid full retraining of the first model for each new input dataset, as a way to improve the method/system of BUCCHOLZ as modified above, by allowing the resulting model to produce a more accurate image reconstruction without sacrificing efficiency and speed.
With respect to Claims 4 and 14, BUCHHOLZ, as modified by PEREZ and MEYERS-NORMAND, as taught above, teaches the limitations of Claims 1 and 11.
BUCHHOLZ further teaches:
wherein generating the second set of data points comprises executing the updated model on a first set of two-dimensional positions within the spectral domain to generate a set of predicted values that correspond to the first set of two-dimensional positions and are associated with the first physical object. (BUCHHOLZ teaches 2D transformation process from original image, FIG. 1, “Fast-Transformers (7), that operate on images via a novel sequential image representation we call Fourier Domain Encoding (FDE)”; and Pg2,2.1: “positional encodings are required whenever specific input topologies need to be made accessible to the transformer. In (5), a useful 1D positional encoding scheme was proposed. Later, Wang et al. (8) generalized this scheme to 2D topologies.”; Examiner interprets data representation of “physical object” as above.)
With respect to Claim 5, BUCHHOLZ, as modified by PEREZ and MEYERS-NORMAND, as taught above, teaches the limitations of Claim 1.
BUCHHOLZ further teaches:
wherein the updated model maps one or more positions within the spectral domain to one or more predicted values associated with both the first physical object and the spectral domain. (BUCHHOLZ teaches mapping positions to predictions, as above, Fig. 2, with Pg1, 1: “we show how an encoder-decoder based Fourier Image Transformer (“FIT: TRec”) can be trained on a set of Fourier measurements and then used to query arbitrary Fourier coefficients, which we use to improve sparse-view computed tomography (CT) image restoration mapping”; BUCHHOLZ teaches predictive modeling and transformation to frequency domain, Pg2,2.3: “first n pixels of the flattened input image are used to condition a generative transformer setup that then predicts the remaining image in an auto-regressive manner”, and Pg.3, Section 3.1, FIG.2 and caption: “Low-resolution input images are first transformed into Fourier space and then unrolled into an FDE sequence…fed to a FIT, that, conditioned on this input, extends the FDE sequence to represent a higher resolution image…FIT is conditioned on the first 39 entries of the FDE”; Examiner notes interpretation of “spectral domain” and “values associated with physical object” as above.)
BUCHHOLZ, does not teach:
the model comprises a neural network that maps one or more positions within the spectral domain to one or more predicted values
MEYERS-NORMAND teaches:
the updated model comprises a neural network within the spectral domain to one or more predicted values (MEYERS-NORMAND teaches use of neural network for generating predictive models, [0041]: “machine learning algorithm can be implemented using a variety of techniques, including the use of one or more an artificial neural network, a deep neural network, a convolutional neural network, a multilayer perceptron, and the like”, and as above, iterative method including spectral domain data for generating predictive models.)
It would have been obvious to one of ordinary skill in the art before effective filing date of the claimed invention to modify BUCHHOLZ, as modified by PEREZ and taught above, to include an updated model that comprises a neural network within the spectral domain to one or more predicted values, as taught by MEYERS-NORMAND, because it would take advantage of the unsupervised training power made possible by a neural network. One of ordinary skill would be motivated to include neural network techniques, as explicitly taught as part of the machine-learning method of MEYERS-NORMAND into the method taught by BUCHHOLZ as modified above to arrive at a more efficient and robust method/system for image reconstruction.
With respect to Claims 6 and 16, BUCHHOLZ, as modified by PEREZ and MEYERS-NORMAND, as taught above, teaches the limitations of Claims 1 and 11.
BUCHHOLZ further teaches:
wherein constructing the image comprises computing an inverse Fourier transform of the second set of data points to generate a third set of data points associated with both the first physical object and a spatial domain. (BUCHHOLZ teaches analogous iterative data sets, Fig.6, and Pg6, 4.6: “Qualitative tomographic reconstruction results for all three datasets we used are shown in Figure 6. For each dataset, we show three input sinograms, the reconstruction baseline obtained via filtered back projection (FBP), our results obtained via “FIT: TRec” and “FIT: TRec + FBP”, and the corresponding ground truth images.” – different inputs yield different reconstruction models”, and teaches technique of using inverse FT, Fig. 4,3.3: “All final prediction images
PNG
media_image1.png
16
14
media_image1.png
Greyscale
are generated by computing the inverse Fourier transform on predictions
PNG
media_image7.png
21
17
media_image7.png
Greyscale
”; Examiner notes “data points association with physical object” as above, analogous to reference.)
With respect to Claims 7 and 17, BUCHHOLZ, as modified by PEREZ and MEYERS-NORMAND, as taught above, teaches the limitations of Claim 1 and 11.
BUCHHOLZ teaches
Image is constructed (BUCHHOLZ teaches, as above, construction of image from trained model,
BUCHHOLZ, as modified by PEREZ and MEYERS-NORMAND, as taught above does not teach:
wherein the image is constructed to have a target level of fidelity.
MEYERS-NORMAND teaches
wherein output is constructed to have a target level of fidelity. (MEYERS- NORMAND teaches “correct” target for result, [0040]: “supervised machine learning algorithm builds a machine learning model by processing training data that includes both input data and desired outputs (e.g., for each input data, the correct answer (also referred to as the “target” or “target attribute”)”; Examiner interprets “target level of fidelity” as analogous to “correct answer” or “target attribute”, as taught in reference.)
It would have been obvious to one of ordinary skill in the art before effective filing date of the claimed invention modify BUCHHOLZ, as modified by PEREZ and MEYERS-NORMAND, as taught above to incorporate a target level of fidelity for output, such as that further disclosed by MEYERS-NORMAND because it would be understood as a way to increase confidence in interpretation of results produced by a generative predictive model. It would be obvious to combine the target attribute comparison as taught by MEYERS-NORMAND into the method/system disclosed by BUCHHOLZ as modified above, to result in the ability to ascertain reliability and accuracy of a reconstructed image. One of ordinary skill would understand that including a reliable fidelity limit or “target attribute, would enhance the ability to balance the competing objectives of image construction, namely accuracy compared to ‘ground truth’, competing with noise level and spatial resolution, among other factors.
With respect to Claim 8, BUCHHOLZ, as modified by PEREZ and MEYERS-NORMAND, as taught above, teaches the limitations of Claim 1.
BUCHHOLZ further teaches:
generating, via the first trained machine learning model, a second model ( As above, BUCHHOLZ teaches iterative model generation with multiple items, different inputs yielding different models for reconstruction, see PG6, 4.6: “For each dataset, we show three input sinograms, the reconstruction baseline obtained via filtered back projection (FBP), our results obtained via “FIT: TRec” and “FIT: TRec + FBP”, and the corresponding ground truth images.”)
BUCHHOLZ, as modified by PEREZ and MEYERS-NORMAND, as taught above does not teach:
generating, via the first trained machine learning model, a second model that represents a second physical based on a third set of data points associated with both the second physical object
MEYERS-NORMAND further teaches:generating, via the first trained machine learning model, a second model that
represents a second physical object based on a third set of data points associated with the second physical object (MEYERS-NORMAND teaches generative of multiple iterative models, as above, FIG. 4, with [0008], and [0009])
It would have been obvious to one of ordinary skill in the art before effective filing
date of the claimed invention to modify BUCHHOLZ, as modified by PEREZ and MEYERS-NORMAND, as taught above to include generating, via the first trained machine learning model, a second model that represents a second item based on a third set of data points associated with both the second item, such as that further disclosed by MEYERS-NORMAND because it would improve the ability to extract meaningful data even from low signal-to-noise acquisition by multiple iterative training steps. One of ordinary skill would be motivated to combine the machine-learning focused steps of MEYERS-NORMAND with the method of image reconstruction using the related Fourier techniques as taught by BUCHHOLZ to make the resulting image models and reconstruction more accurate and reliable.
With respect to Claims 9 and 19, BUCHHOLZ, as modified by PEREZ and MEYERS-NORMAND, as taught above, teaches the limitations of Claim 1.
BUCHHOLZ, as modified by PEREZ and MEYERS-NORMAND, as taught above does not teach:
wherein the first physical object comprises an astronomical object, a body organ, or a surface
MEYERS-NORMAND teaches:
wherein the first physical object comprises an astronomical object, a body organ, a surface (MEYERS-NORMAND teaches analysis and image reconstruction of retinal surface, see Abstract: “patient data corresponding to the retinal images”, which is both a body organ and a surface structure, and FIG. 5 )
It would have been obvious to one of ordinary skill in the art before effective filing
date of the claimed invention to modify BUCHHOLZ, as modified by PEREZ and MEYERS-NORMAND, as taught above to include a first physical object comprising an astronomical object, a body organ, or a surface, as further taught by MEYERS-NORMAND because this would broaden the range of applications of the method.
With respect to Claim 15, BUCHHOLZ, as modified by PEREZ and MEYERS-NORMAND, as taught above, teaches the limitations of Claim 11.
BUCHHOLZ further teaches:
wherein the first trained machine learning model comprises at least one of a transformer encoder, a variational encoder, or a learnable neural spine. (BUCHHOLZ is directed to use of Fourier Image transformer, see Title; based on encoding/decoding, see Pg1,1: “We demonstrate this by providing a given set of projection Fourier coefficients to our encoder-decoder setup and use it to predict Fourier coefficients at arbitrary query points.”)
With respect to Claim 18, BUCHHOLZ, as modified by PEREZ and MEYERS-NORMAND, as taught above, teaches the limitations of Claim 11.
BUCHHOLZ further teaches:
wherein the spectral domain comprises a frequency domain, a k-space, a cepstral domain, or a wavelet domain. (BUCHHOLZ teaches, as above, Fourier methods, with Fourier transforms to frequency domain, Pg.1, FIG. 1, w/caption: “Fourier Image Transformers (FITs), realizations of Fast-Transformers (7), that operate on images…Fourier Domain Encoding (FDE) …second task is tomographic reconstruction (bottom row), where a given set of projection images (a sinogram) is transformed by an encoder-decoder FIT to improve the quality of the reconstructed image transformation (iFFT)”)
Claim 10 is rejected under pre-AIA 35 U.S.C. 103(a) as being unpatentable over BUCHHOLZ, in view of PEREZ and MEYERS-NORMAND, as applied above to Claim 1, and further in view of HONMA (Honma, et al., "Super-Resolution Imaging with Radio Interferometry using Sparse Modeling", Publications of the Astronomical Society of Japan, Vol. 66, No.5, 2014)
With respect to Claim 10, BUCHHOLZ, as modified by PEREZ and MEYERS-NORMAND, as taught above, teaches the limitations of Claim 1.
BUCHHOLZ, as modified by PEREZ and MEYERS-NORMAND, as taught above does not teach:
wherein the first set of data points comprises an interferometric observation of the first physical object.
HONMA teaches:
wherein the first set of data points comprises an interferometric observation of the first physical object. (HONMA is in same technical field or image reconstruction, directed toward astronomical images, Abstract: “new technique to obtain super-resolution images with radio interferometer using sparse modeling” and Pg2,1.2: “different approaches to reconstruct radio interferometer images”; HONMA teaches interferometric image data for analysis, Abstract: “present results of one-dimensional and two-dimensional simulations of interferometric imaging,”; HONMA teaches image analysis for reconstruction of interferometric data, Pg2,2.1: “imaging synthesis, side-lobe levels can be reduced by using taper function instead of equal weight to all the sampled data.”; and Pg6-7,4.3see Fig.4: “image reconstruction is done in the same manner using LASSO”; also see Fig. 6. “Imaging results for simulated EHT observations of M87’s black hole shadow” (i.e., “physical object”))
It would have been obvious to one of ordinary skill in the art before effective filing date of the claimed invention to further modify BUCHHOLZ, as modified by PEREZ and MEYERS-NORMAND as taught above, to include the first set of data points comprises an interferometric observation of the first item, as taught by HONMA because this would be a useful and important application of the method/system of image reconstruction disclosed by BUCHHOLZ, and modified by PEREZ and MEYERS-NORMAND as modified above. One of ordinary skill would be motivated to use the teaching of HONMA to broaden the range of use for a machine learning-based image reconstruction method/system by including the option of interferometric data would broaden the application range. One of ordinary skill would realize the advantage of combining the high resolution spatial frequency information of interferometric data with trained machine learning algorithms designed to work in a spectral/frequency domain as taught by BUCHHOLZ, as modified above.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
TANCIK (Tancik, et al., “Fourier Features Let Networks Learn High Frequency Functions in Low Dimensional Domains”, arXiv:2006.10739v1 [cs.CV] 18 Jun 2020) – teaches imaging and reconstruction of physical objects explicitly, relevant to all claims in the instant application; uses Fourier methods.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to TONI D SAUNCY whose telephone number is (703)756-4589. The examiner can normally be reached Monday - Friday 8:30 a.m. - 5:30 p.m. ET.
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, Catherine Rastovski can be reached at 571-270-0349. 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.
/TONI D SAUNCY/Examiner, Art Unit 2857
/Catherine T. Rastovski/Supervisory Primary Examiner, Art Unit 2857
Prosecution Insights