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and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114.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 August 12th, 2026 has been entered.
Response to Amendment
The Amendment filed on August 12th, 2026 has been entered. Claims 1, 6–9, and 14–16 are currently pending. Claims 1, 6, 9, and 14 have been amended, and Claims 2–5 and 10–13 are canceled.
Response to Arguments
Applicant’s arguments filed August 12th, 2026, concerning the rejection under 35 U.S.C. § 112(a) have been fully considered and are persuasive. Amended independent claims 1 and 9 now limit the generation model to learning melanoma features and nevus features based on specified morphological differences between melanoma and nevus. The specification describes the melanoma and nevus learning data, a GAN or StyleGAN2 generation model, generator-and-discriminator adversarial training, and example melanoma images paired with corresponding generated images. Accordingly, upon reconsideration of the record as a whole, the previous rejection of claims 1, 6–9, and 14–16 under 35 U.S.C. § 112(a) is withdrawn.
Applicant’s arguments filed August 12, 2026, concerning the prior rejections under 35 U.S.C. § 103 have been fully considered and are persuasive to the extent that the previously applied combinations do not fully address the morphological-difference-based generation-model limitation and the predefined-number pixel-selection limitation now incorporated into independent claims 1 and 9. Accordingly, the previous rejection of claims 1, 7–9, and 15–16 under 35 U.S.C. § 103 as unpatentable over Lapiere in view of Lee is withdrawn. The previous rejection of claims 4, 6, 12, and 14 under 35 U.S.C. § 103 as unpatentable over Lapiere in view of Lee and further in view of Teixeira is also withdrawn. However, new grounds of rejection under 35 U.S.C. § 103 are made in this Office action applying Zunair in view of Lapiere and Liang to presently pending claims 1, 6–9, and 14–16, as set forth in detail below.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claim(s) 1, 6–9, and 14–16 are rejected under 35 U.S.C. 103 as being unpatentable over Zunair (Zunair et al. (2020). Melanoma detection using adversarial training and deep transfer learning. Physics in Medicine & Biology, 65(13), 135005) in view of Lapiere (Lapiere et al, US 2019/0340762 A1, 2019), further in view of Liang (Liang et al, US 2022/0084173 A1, 2022).
Regarding claim 1, Zunair discloses an electronic device for AI-based recommendation of a melanoma biopsy site ( [Page. 1, Abstract and §1 “Introduction”], [Page. 2, §1 “Introduction”], [Page. 5, “Implementation details”]: Zunair teaches a computer-aided diagnostic system for automatic classification and early detection of melanoma from dermoscopic skin images. Zunair’s MelaNet is a deep-neural-network framework for melanoma detection that performs adversarial training, image generation, and skin-lesion classification. ), comprising:
a processor configured to:
( [Page. 5, “Implementation details”], [Page. 4, §2.4 and Algorithm 1], [Pages. 2–3, §2.1]: Zunair teaches that the disclosed MelaNet algorithms are implemented in Keras with a TensorFlow backend and that all experiments are performed on a Linux server having two Intel Xeon E5-2650 V4 processors, 256 GB RAM, and four NVIDIA P100 GPUs. Algorithm 1 identifies the CycleGAN training, image translation, VGG-GAP training, and classification operations executed by the computer system. )
classify a skin image as melanoma or nevus by inputting the skin image to a classification model,
( [Page. 3, §2.2 “Pre-trained Model Architecture” and Fig. 2], [Page. 4, §2.4 “Algorithm” and Algorithm 1], [Page. 6, §3.1 “Results”]: Zunair teaches a VGG-GAP convolutional-neural-network classification model that receives a dermoscopic skin-lesion image and outputs, through a two-unit softmax layer, probabilities for the benign [nevus] and malignant [melanoma] classes. Algorithm 1 specifies inputting dermoscopic images into the trained VGG-GAP model and generating predicted class labels. Zunair further characterizes the binary classification task as distinguishing melanoma from non-melanoma images and reports classification of skin-lesion images as benign or malignant.)
when the skin image is classified as melanoma, identify melanoma features in the skin image by using a generation model to generate an image from which the melanoma features are removed,
( [Page. 2, §2.1 “Conditional Image Synthesis”], [Page. 3, §2.1, Fig. 1 and Equation 3], [Page. 4, §2.4 and Algorithm 1]: Zunair teaches classifying dermoscopic skin-lesion images as malignant or benign and teaches a CycleGAN having a generator (GM : M →B), where (M) represents the malignant domain and (B) represents the benign domain. Zunair expressly states that (GM) receives a malignant image as input and generates a realistic benign lesion image [melanoma features are removed], and Fig. 1 illustrates image translation in the backward direction from malignant to benign. Under the construction that malignant represents melanoma and benign represents nevus, applying (GM) to an image classified as malignant generates a corresponding benign image from which the malignant or melanoma features are removed. )
wherein the generation model is learned to identify the melanoma features and nevus features based on a morphological difference between melanoma and nevus, the morphological difference including at least one of a shape, a pigmentation pattern, a pigmentation pattern in ridges and furrows, or a color;
( [Page. 1, Abstract and §1 “Introduction”], [Page. 4, §2.4 “Algorithm” and Algorithm 1], [Pages. 5 and 7–8, “Dataset,” “Discussion,” and §4 “Conclusion”]: Zunair teaches training CycleGAN on benign and malignant dermoscopic-image domains to learn an inter-class mapping and a function of the inter-class variation, which Zunair describes as a transformation between melanoma and non-melanoma lesions. Zunair further teaches that image-to-image translation adds or removes image features and identifies color, texture and shape as visual characteristics or variations [morphological difference] associated with melanoma lesions. )
determine pixel values according to color of pixels of the skin image and the generated image, the pixel values identified as three-dimensional coordinate values having a first-channel value, a second-channel value, and a third-channel value as axes, respectively, according to respective values of the color of each pixel;
( [Page. 1, Abstract and §1 “Introduction”], [Page. 3, §2.1, Fig. 1, and Eq. (3)], [Page. 4, §2.4 and Algorithm 1]: Zunair teaches processing color dermoscopic skin-lesion images and teaches that generator (GM) receives a malignant image (m) and generates a realistic benign image (GM(m)). Zunair further teaches resizing each image to 256 × 256 × 3 and combining the generated target-class images with the original training images. A person of ordinary skill in the art would have understood from the 256 ×256 ×3 image format that each pixel of the original and generated images is represented by three respective image-channel values corresponding to the color of that pixel. )
calculate a distance between pixels for each pixel by using respective channel values of corresponding pixels of the skin image and the generated image,
( [Page. 3, §2.1, Fig. 1, and Eq. (3)], [Page. 3, Eq. (4)], [Page. 4, §2.4 and Algorithm 1]: Zunair teaches generating a benign-domain image (GM(m)) from a malignant input image (m), resizing each dermoscopic image to 256 × 256 × 3, and applying an ℓ1-norm to the difference between an input image and a generated cycle-reconstructed image. Accordingly, the subtraction of the conformable 256 × 256 × 3 image tensors uses the respective channel values at corresponding pixel positions to calculate the ℓ1 image difference. )
Zunair supplies three-channel pixel data for both of the skin image and the generated image. Lapiere identifies those channels specifically as R, G, and B coordinate axes:
determine pixel values according to color of pixels of the skin image, the pixel values identified as three-dimensional coordinate values having R (Red), G (Green), and B (Blue) as axes, respectively, according to RGB values of the color of each pixel;
( [0005], [0081–0082]: Lapiere teaches pixel-value computations for skin-image analysis and teaches placing pixels in a three-dimensional space, wherein the three axes represent the R-G-B color channels, and further teaches computing three-dimensional Euclidean coordinate values/metrics in that RGB space. )
It would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to apply Lapiere’s known three-dimensional RGB representation and color-distance technique to corresponding pixels of Zunair’s three-channel dermoscopic input and generated images. Because both references analyze skin-lesion color, the modification would have predictably quantified pixel-color changes introduced by Zunair’s melanoma-to-benign translation, thereby facilitating localization of the removed melanoma features without altering Zunair’s CycleGAN operation.
Zunair [as modified by Lapiere] teaches corresponding-pixel RGB distances for the original melanoma image and generated benign image. Liang further teaches using such pixelwise image differences to generate a lesion-localization map and superimpose the localized region on the original image:
calculate a distance between pixels for each pixel by using coordinate values of corresponding pixels of the skin image and the generated image,
( [0032], [0047 and Fig. 2], [0106–0112 and Figs. 6A–6C]: Liang teaches translating a diseased medical image into a healthy image and subtracting the translated output image from the input image to generate a difference map. Liang further teaches evaluating values across all pixels of the difference map, binarizing the map, and clustering its foreground pixels for lesion localization; thereby subtracting conformable input and translated images to produce a spatial, pixel-valued map compares corresponding pixel positions. )
identify a predefined number of pixels as at least one candidate biopsy site based on the distance between pixels, and
( [0034], [0043–0044], [0111], [0112]: Liang teaches identifying a minimal subset of changed pixels, evaluating the maximum value across the pixels of the difference map, binarizing the difference map based on its pixel values, clustering the foreground pixels into connected components, and treating each connected component having an area greater than ten pixels as a lesion candidate. )
display the at least one candidate biopsy site on the skin image.
( [0106–0107], [0110], [Figs. 6A–6C, localization columns]: Liang teaches that differences between a diseased input image and its generated healthy image reveal disease locations and expressly teaches superimposing the localized diseased regions on the original images. Liang further teaches visualizing the difference maps for lesion localization. )
It would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Zunair [as modified by Lapiere] by applying Liang’s input/ output difference-map localization technique to the corresponding three-dimensional RGB pixel coordinates of Zunair’s melanoma image and generated benign image, selecting a predefined number of pixels having the greatest resulting color distances as candidate biopsy sites, and superimposing the selected locations on the original skin image. Liang teaches that differences between a diseased input image and its generated healthy image reveal disease locations, evaluates difference-map values across all pixels, identifies foreground pixel components as lesion candidates, and superimposes the localized disease regions on the original image. Zunair [as modified by Lapiere] teaches representing skin-image pixels in three-dimensional RGB space, calculating distances in that space, and selecting pixels associated with the greatest computed distances. A person of ordinary skill in the art therefore would have been motivated to use the greatest corresponding-pixel RGB distances to prioritize a predefined, manageable number of locations most affected by Zunair [as modified by Lapiere]’s melanoma-to-benign translation. Selecting a fixed number of the highest-distance pixels instead of retaining a variable number of thresholded pixels would have been a predictable candidate-limiting technique that reduces display clutter and presents the clinician with the locations most likely to contain the melanoma features removed by the generation model. The modification would have predictably resulted in identifying those locations as candidate biopsy sites and displaying them on Zunair [as modified by Lapiere]’s original dermoscopic skin image, with a reasonable expectation of success because Liang’s difference-map localization and visualization operations are performed on spatially corresponding input and generated images.
Regarding claim 6, Zunair [as modified by Lapiere and Liang] teaches the electronic device of claim 1, wherein the processor is configured to use the skin image and the generated image by filtering with a Gaussian filter.
( [0060 and Fig. 7], [0061 and Fig. 8], [0062 and Fig. 8]: Lapiere teaches creating, from an enhanced RGB skin image, a Laplacian-of-Gaussian pyramid comprising a collection of images filtered using successive Laplacian-of-Gaussian filters. Lapiere further teaches converting the RGB skin image to grayscale and applying a series of Laplacian-of-Gaussian filters having different standard deviations and sizes to form the filtered-image pyramid and highlight skin moles or anomalies. )
Regarding claim 7, Zunair [as modified by Lapiere and Liang] teaches the electronic device of claim 1, wherein the classification model is learned to classify whether a skin image being input is melanoma or nevus by using, as learning data, a plurality of skin images including a plurality of melanoma images and a plurality of nevus images and answer information on whether each skin image is melanoma or nevus.
( [Page 3, §2.2 “Pre-trained Model Architecture” and Fig. 2], [Page 4, §2.4 “Algorithm” and Algorithm 1], [Pages 4–5, §3 “Dataset,” “Training Procedure,” and “Implementation details”]: Zunair teaches a VGG-GAP classification model having a two-unit softmax output for binary benign/malignant classification. Algorithm 1 defines the training set as a plurality of dermoscopic images (D={(I1,y1),…,(In,yn)}), wherein each (yi) is the class label associated with the respective input image, groups each lesion image according to its class label, and trains VGG-GAP on the balanced training set. Zunair further teaches that the ISIC-2016 training set contains 727 benign images and 173 malignant images and that the complete VGG-GAP network is trained using the labeled images by minimizing the focal loss. )
Regarding claim 8, Zunair [as modified by Lapiere and Liang] teaches the electronic device of claim 1, wherein the generation model is learned to identify melanoma features and nevus features from a plurality of skin images including a plurality of melanoma images and a plurality of nevus images
( [Pages 2–3, §2.1 “Conditional Image Synthesis,” Fig. 1, and Equations 1–3], [Page 4, §2.4 “Algorithm” and Algorithm 1, lines 2 and 5–6], [Pages 5 and 7, “Dataset” and “Discussion”]: Zunair teaches partitioning a plurality of dermoscopic skin-lesion images into benign and malignant image domains and training a bidirectional CycleGAN on unpaired images from the two domains. Zunair teaches that CycleGAN learns mappings between the benign and malignant domains and a function of the inter-class variation between the two groups. Zunair further teaches that image-to-image translation adds or removes image features. )
and to generate an image in which the melanoma features are removed from the melanoma images.
( [Page 2, §2.1 “Conditional Image Synthesis”], [Page 3, §2.1, Fig. 1, and Equation 3], [Page 7, “Discussion”]: Zunair teaches generator (GM: M → B), which translates images from the malignant domain to the benign domain. Zunair expressly teaches that (GM) receives a malignant image (m) and generates a realistic image (GM(m)) in the benign domain, with the objective of generating a benign lesion, and Fig. 1 depicts the backward translation from malignant to benign. Zunair further teaches that image-to-image translation removes image features. )
Regarding claims 9 and 14–16. The rationale provided for claims 1 and 7–8 is incorporated herein. In addition, the electronic device of claims 1 and 7–8 correspond to the method of claims 9 and 15–16. Therefore, the claims are all rejected.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to KEN KUDO whose telephone number is (571)272-4498. The examiner can normally be reached M-F 8am - 5pm.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Vincent Rudolph can be reached at 571-272-8243. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
KEN KUDO
Examiner
Art Unit 2671
/KEN KUDO/Examiner, Art Unit 2671
/VINCENT RUDOLPH/Supervisory Patent Examiner, Art Unit 2671
Prosecution Insights