Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Priority
Receipt is acknowledged of papers submitted under 35 U.S.C. 119(a)-(d), which papers have been placed of record in the file.
Information Disclosure Statement
The information disclosure statements (IDS) submitted on 03/21/2025 and 12/03/2025 are in compliance with the provisions on 37 CFR 1.97. Accordingly, the information disclosure statements are being considered by the examiner.
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.
Claims 1, 2, 6, 12, 13 and 14 are rejected under 35 U.S.C. 103 as being unpatentable over Yang (CN 113288186 A) in view of Nakayama (US 2020/0146645 A1) and further in view of Agarwal et al., “Deep learning for mass detection in Full Field Digital Mammograms,” Computers in Biology and Medicine (2020) (hereinafter referred to as Agarwal).
Regarding claim 1, Yang teaches an information processing apparatus (Yang, page 14, , Fig 3, breast tumor detection device: A breast tumour tissue detecting device based on deep learning algorithm, comprising: a memory and a processor, the memory storing a computer program, the processor, when executing the computer program, implementing the method for detecting breast tumour tissue based on deep learning algorithm according to any one of claims 1 to 8.) comprising: at least one processor, wherein the processor is configured to (Yang, claim 9): generate a difference image representing a difference between a low-energy image captured by irradiating a subject, in which a contrast agent is injected, with electromagnetic waves having first energy and a high-energy image captured by irradiating the subject with electromagnetic waves having second energy higher than the first energy (Yang, pages 8-9, Description of pictures, Fig1, Step 1.1, Yang teaches injecting an iodine-containing contrast agent into the patient before imaging, capturing a low-energy (LE) image of the breast, capturing a high-energy (HE) image of the breast alongside the LE image and generating a subtraction (difference) image from the LE and HE images using the K-edge effect of the iodine contrast agent.); extract a lesion candidate region including a lesion candidate from the difference image (Yang, pages 9-11, Description of pictures, Step 2: Yang teaches establishing a separate deep learning model specifically for the subtraction (difference) image., Step 2.2, claim 6, Yang teaches using a ResNet network to extract features from the subtraction image, forming a feature pyramid for multi-scale detection., Step 2.3; Claim 6, Yang teaches establishing boundary frame regression sub-networks on each layer of the feature pyramid to predict bounding box coordinates, thereby identifying candidate regions in the subtraction image., Step 2.4; Claim 6, Yang teaches extracting the focus image (i.e., the lesion candidate region) from the segmented subtraction image.); cut out a region corresponding to the lesion candidate region from at least any one of the low-energy image or the high-energy image, as a patch image (Yang, pages 9-11, Description of pictures, Step 2: Yang teaches two separate deep learning models operating in parallel (one for the LE image and one for the subtraction image). Each model detects, extracts, and classifies within its own image type.
However, Yang does not explicitly teach cutting out a region from the LE image at a location identified in the subtraction (difference) image (the cross-image-type patch image).
Reference Nakayama teaches a mammography system that performs dual-energy subtraction imaging with an iodine contrast agent (Nakayama, paragraphs [0063]-[0065]). Nakayama teaches detecting a region of interest (ROI) from differential tomographic images (Nakayama, paragraph [0145]: “detects a region of an interesting object (ROI) from the plurality of differential tomographic images”), deriving the three-dimensional position of the detected ROI (Nakayama, paragraph [0146]), and then deriving a region corresponding to the ROI from the LE-based tomographic images based on the position detected in the differential image (Nakayama, paragraph [0147]: “derives a region corresponding to the region of the ROI from the plurality of tomographic images generated in step S304 on the basis of the position of the ROI derived in step S312”). This teaches the cross-image-type concept: detect in the difference image, then derive (cut out) the corresponding region in the LE-based image.
These arts are analogous since they are both related to medical imaging devices. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention (AIA ) to modify Yang’s parallel-pipeline system by applying Nakayama’s cross-image-type approach detecting the lesion in the subtraction image, then extracting the corresponding region from the LE image for classification because:
Nakayama demonstrates that detecting a lesion in the differential image and deriving the corresponding region in the LE-based image is a known, effective approach (Nakayama, paragraphs [0145]-[0149]).
Yang recognizes that both image types provide complementary information (Yang, Abstract: “combining low energy image and high and low energy subtraction image…making up the deficiency of single low-energy image detection”). The subtraction image is superior for detection; the LE image provides structural detail for classification.
Agarwal demonstrates that two-stage detection architectures (detect candidate regions classify each region) were the dominant paradigm for mammography mass detection before the effective filing date, achieving state-of-the-art results (Agarwal, Abstract: TPR 0.93 at 0.78 FPI on approximately 80,000 FFDMs). A person of ordinary skill in the art would recognize the modification as a natural adaptation of this established two-stage paradigm to CEDM imaging.
The modification amounts to combining familiar elements according to known methods to yield predictable results. See KSR Int’l Co. v. Teleflex Inc., 550 U.S. 398, 416 (2007).
Regarding claim 2, the combination of Yang in view of Nakayama (US and further in view of Agarwal teaches the information processing apparatus according to claim 1, and wherein the processor is configured to: determine whether or not the lesion candidate is the lesion by inputting the patch image to a machine learned model (Yang, pages 9-11, Description of pictures, Step 2.2, claim 6, Yang teaches using deep learning models, specifically ResNet feature extraction networks and CNN-based classification networks, for lesion detection and benign/malignant classification. ResNet and CNN classifiers are machine learned models.).
Regarding claim 6, the combination of Yang in view of Nakayama (US and further in view of Agarwal teaches the information processing apparatus according to claim 1, and wherein the subject is left and right breasts (Yang, pages 8-12, Description of pictures, Steps 1.1, 3.1, Yang teaches imaging both left and right breasts), the low-energy image includes a first low-energy image and a second low-energy image that are captured by irradiating each of the left and right breasts with radiation having the first energy, the high-energy image includes a first high-energy image and a second high-energy image that are captured by irradiating each of the left and right breasts with radiation having the second energy, and the difference image includes a first difference image representing a difference between the first low-energy image and the first high-energy image and a second difference image representing a difference between the second low-energy image and the second high-energy image (Yang, pages 8-12, Description of pictures, Steps 1.1, 3.1, Yang teaches obtaining LE images, HE images, and subtraction images for each breast at multiple positions (Steps 1.1,1.2, 3.1). Since Yang captures LE and HE images for each breast and generates subtraction images for each, Yang inherently generates separate difference images for the left and right breasts.). Nakayama likewise teaches imaging both left and right mammae (Nakayama, paragraph [0051]).
Regarding claim 12, the combination of Yang in view of Nakayama (US and further in view of Agarwal teaches the information processing apparatus according to claim 2, and wherein the machine learned model is generated by training a machine learning model using training data including an input image and ground-truth data and an augmented image in which a contrast between a lesion region and a non-lesion region in the input image is changed (Yang, claim 7, Yang teaches continuously updating the deep learning model by adding new confirmed cases).
Agarwal teaches using data augmentation during training of the mammography detection model, including image transformations (Section 4.1: “horizontal flipping is applied as data augmentation”; Section 5.3: augmented dataset with rotation). Agarwal further demonstrates that contrast variation between different scanner manufacturers is a known challenge requiring image normalization (Sections 2.4, 6).
Contrast augmentation, adjusting brightness and contrast of training images to improve model robustness, is a well-known, standard data augmentation technique in the deep learning art.
It would have been obvious to include contrast-variation augmentation because (1) Yang’s system processes LE and HE images whose contrast naturally varies depending on contrast agent uptake, breast density, and imaging parameters; and (2) Agarwal recognizes that contrast differences between imaging systems affect model performance, making contrast augmentation a predictable training strategy.
Claims 13 and 14 are rejected for the same reasons as claim 1 above.
Claim 3 is rejected under 35 U.S.C. 103 as being unpatentable over Yang (CN 113288186 A) in view of Nakayama (US 2020/0146645 A1) and Agarwal et al., “Deep learning for mass detection in Full Field Digital Mammograms,” Computers in Biology and Medicine (2020) (hereinafter referred to as Agarwal) as applied to claim 2 above and further in view of Tanaka (JP 2020010805 A) and Xu, et al. (US 6240201 B1) (hereinafter referred to as Xu).
Regarding claim 3, as set forth in the rejection of claims 1 and 2 above, the combination of Yang in view of Nakayama and Agarwal teaches extracting a lesion candidate from the difference image and cutting a patch from the LE image at the candidate location for ML-based classification.
However, Yang, Nakayama and Agarwal do not teach providing patches from both the LE image and the difference image to the same ML model.
The reference Tanaka teaches, in Embodiment 4, generating a difference image by subtracting bilaterally corresponding regions of left and right breast mammography images (Tanaka, paragraphs [0094]-[0096]; FIG. 27). Tanaka teaches extracting a tumor region from the mammography image and classifying using a first CNN classifier, and extracting a tumor region from the difference image at the same location and classifying using a second CNN classifier (Tanaka, paragraphs [0097]–[0099]). Tanaka teaches integrating both classification results to achieve improved discrimination accuracy (Tanaka, paragraphs [0102]; [0043]: 66.7% integrated vs. 64.6% and 62.5% individually). This teaches extracting from both the structural image and the difference image at the same tumor location but using separate classifiers rather than a single model.
The reference Xu teaches a dual-energy CAD system for lung nodule detection. Xu teaches extracting features from both the standard digital chest image AND its difference image for each abnormal candidate, and supplying both sets of features to the same artificial neural network (Col 4, 5, Steps 40, 50: “for each of the remaining nodule candidates, the features extracted from both the standard chest image and its difference image in step 40 are supplied as ANN inputs”). This teaches providing multi-source features from both the original image and the difference image to a single neural network for classification.
It would have been obvious to additionally cut a patch from the difference image at the same candidate location and provide both patches to the same ML model because:
Tanaka demonstrates that classifying from both the structural image and the difference image at the same location improves accuracy (Tanaka, paragraph [0043]).
Xu teaches that features from both the original image and the difference image can be supplied to the same neural network (Xu, Steps 40/50).
Yang recognizes the complementary value of both image types (Yang, Abstract: “combining low energy image and high and low energy subtraction image…making up the deficiency of single low-energy image detection”).
Providing both patches as multi-channel input to a single CNN is a standard technique in the deep learning art for combining complementary image representations.
Claim 7 is rejected under 35 U.S.C. 103 as being unpatentable over Yang (CN 113288186 A) in view of Nakayama (US 2020/0146645 A1) and Agarwal et al., “Deep learning for mass detection in Full Field Digital Mammograms,” Computers in Biology and Medicine (2020) (hereinafter referred to as Agarwal) as applied to claim 6 above and further in view of Tanaka (JP 2020010805 A)
Regarding claim 7, as set forth in the rejection of claim 6, Yang in view of Nakayama and Agarwal teaches generating separate difference images for left and right breasts and extracting lesion candidates using deep learning.
However, Yang, Nakayama and Agarwal do not explicitly teach separate dedicated models for the left and right breast difference images.
Reference Tanaka teaches separate, dedicated classifiers for different inputs, a first classifier trained on a first image type and a second classifier trained on a second image type (Tanaka, paragraph [0006]; Claim 1). Tanaka experimentally demonstrates that separate dedicated classifiers with integrated results achieve higher accuracy than a single shared classifier (Tanaka, paragraph [0043]: 66.7% integrated vs. 58.3% combined two-input).
It would have been obvious to use separate dedicated models for the left and right breast difference images because:
Tanaka teaches that separate dedicated classifiers with integrated results achieve higher accuracy than a shared classifier (Tanaka, paragraph [0043]).
Bilateral comparison is a fundamental clinical principle in mammography — radiologists routinely compare left and right breast images. Separate dedicated models allow each to specialize.
Yang already images both breasts and generates separate subtraction images for each (Steps 1.1, 3.1).
Providing separate models is a straightforward application of Tanaka’s architecture.
Allowable Subject Matter
Claims 4-5 and 8-11 are objected to as being dependent upon a rejected base claims but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
Claim 4 recites the limitation “determine an enhancement level of background mammary gland parenchyma of the breast based on the difference image and adjust a parameter for determining a condition under which the machine learned model extracts the lesion candidate region, based on the enhancement level”. In essence the processor acquires a background parenchymal enhancement (BPE) level of the subject and selects the machine learned model from among a plurality of machine learned models based on the BPE level.
The prior art of record does not teach or suggest this limitation.
Yang teaches CEDM-based breast tumor detection using deep learning but does not teach acquiring BPE levels or selecting models based on BPE. Nakayama teaches dual-energy subtraction imaging with ROI detection but does not teach BPE-based model adaptation. Tanaka teaches multi-classifier architectures but does not teach model selection based on imaging characteristics such as BPE. Shu teaches dual-energy CAD for lung nodules but does not address BPE. Agarwal teaches Faster R-CNN for mammography mass detection but does not address BPE or patient-specific model selection.
The concept of maintaining a plurality of ML models optimized for different BPE levels and selecting the appropriate model based on the subject’s actual BPE represents a specific, non-obvious adaptation of the detection pipeline to account for inter-patient variability. No prior art reference of record, individually or in combination, teaches, suggests, or makes obvious this BPE-adaptive model selection.
Claim 5 depends on claim 4 and is therefore allowed for the same reasons as claim 4 above.
Claim 8 recites the additional limitations “wherein the first machine learned model includes a first pre-stage operation block and a first post-stage operation block, the second machine learned model includes a second pre-stage operation block and a second post-stage operation block, and the processor is configured to: extract a first feature value by inputting the first difference image to the first pre-stage operation block; extract a second feature value by inputting the second difference image to the second pre-stage operation block; extract the lesion candidate region from the first difference image by combining the second feature value with the first feature value and inputting the combined feature value to the first post-stage operation block; and extract the lesion candidate region from the second difference image by combining the first feature value with the second feature value and inputting the combined feature value to the second post-stage operation block”. This constitutes a bidirectional cross-branch feature fusion architecture in which each branch’s post-stage processing is informed by the other branch’s pre-stage features. The pre-stage blocks operate independently (parallel feature extraction), but the post-stage blocks operate on cross-combined features each branch receiving complementary information from the other branch before performing its final lesion extraction.
The prior art of record, whether considered individually or in any reasonable combination, does not teach or suggest the specific dual-branch, cross-combination architecture recited in claim 8, taken in combination with all limitations of claims 1, 6, and 7.
Claims 9-10 depend on claim 8 and are allowed for the same reasons as claim 8 above.
Claim 11 recites the additional limitations “wherein the processor is configured to: determine a first enhancement level of background mammary gland parenchyma based on the first difference image; determine a second enhancement level of the background mammary gland parenchyma based on the second difference image; determine symmetry of enhancement regions of the background mammary gland parenchyma related to the left and right breasts based on the first difference image and the second difference image; adjust a parameter for determining a condition under which the first machine learned model extracts the lesion candidate region, based on the first enhancement level and the symmetry; and adjust a parameter for determining a condition under which the second machine learned model extracts the lesion candidate region, based on the second enhancement level and the symmetry”. In essence, the processor acquires a first BPE level of a first breast and a second BPE level of a second breast and uses a first machine learned model corresponding to the first BPE level for the first difference image and a second machine learned model corresponding to the second BPE level for the second difference image.
The prior art of record does not teach or suggest this limitation.
Claim 11 extends the BPE-adaptive concept of claim 4 to a bilateral configuration acquiring BPE levels independently for each breast and selecting separate BPE-appropriate models for each breast’s difference image.
None of the prior art references individually or in combination, teach, suggest, or make obvious: (i) acquiring BPE levels for each breast independently; (ii) maintaining a plurality of models corresponding to different BPE levels; or (iii) selecting separate BPE-appropriate models for the left and right breast difference images.
Related Art
The following cited references were identified during the search and are made of record as pertinent to the applicant’s disclosure, although they are not relied upon for any rejection:
Machii (US 2025/0182278 A1) -This co-pending application by the same inventor. Both applications recognize the same problem and use the same solution concept: Both specifications explicitly recognize that the difference image alone is insufficient for accurate lesion detection because it lacks mammary gland structural information. The difference is when and how the low-energy (LE)/(high-energy (HE) information is introduced: simultaneously (US 2025/0182278 A1) versus sequentially at the classification stage (instant application). Also the instant application extracts a lesion candidate region including a lesion candidate from the difference image; cut out a region corresponding to the lesion candidate region from at least any one of the low-energy image or the high-energy image, as a patch image; and determine whether or not the lesion candidate is a lesion based on the patch image.
Machii (US 2025/0186011 A1) -This co-pending application by the same inventor addresses CEDM image interpretation by independently detecting lesion regions from both the low-energy image and the difference image using separate ML models and performing comparison determination of detection positions. While this application shares the same imaging modality and image types as the present application, the claims are directed to a fundamentally different architecture (parallel dual-model comparison pipeline) and address a different clinical need (BI-RADS reporting attribution of findings to specific image types).
Wang (EP 3926537 B1) -This reference teaches medical image segmentation using temporal dynamic images extracted from time-series DCE-MRI data via Clifford algebra, with a deep learning segmentation model. While the reference teaches the general concept of generating a processed image from multi-source medical images and performing ML-based segmentation, it does not teach dual-energy imaging, difference image generation, or the claimed two-stage decoupled pipeline with patch-based classification from a different image type.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SUPERVISORY PATENT EXAMINER TWYLER HASKINS whose telephone number is (571)272-7406. The SUPERVISORY PATENT EXAMINER TWYLER HASKINS can normally be reached Mon- Thursday: 7:30 am-4:30 pm.
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:/Awwww.uspto.gov/interviewpractice.
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:/Awww.uspto.gov/patents/apply/patent-center for more information about Patent Center and https:/Awww.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.
/TWYLER L HASKINS/ Supervisory Patent Examiner, Art Unit 2639
Prosecution Insights