DETAILED ACTION
This action is in response to Applicant’s Request for Continued Examination ("Response”) received on December 22, 2025, in response to the Office Action dated September 11, 2025. This action is made Non-Final.
Claims 54, 57, 65, 67-69, 71-73, 77, 78, 80, 82, 83, and 90-96 are pending in the case.
Claims 54, 90, and 93 are independent claims.
Claims 54, 57, 65, 67-69, 71-73, 77, 78, 80, 82, 83, and 90-96 are rejected.
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 .
Applicant’s Response
In Applicant’s Response, Applicant amended claims 54, 90, and 93, added claim 96, and submitted arguments against the prior art in the Office Action dated September 11, 2025.
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.
Claim(s) 54, 57, 65, 67-69, 71, 73, 77, 78, and 90-95 is/are rejected under 35 U.S.C. 103 as being unpatentable over Kisilev et al., US Publication no. 2018/0247405 (“Kisilev”), in view of Assmann, US Publication no. US 2008/0170769 (“Assmann”), and further in view of Abedini et al., US Publication 2018/0122076 (“Abedini”).
Claim 54:
Kisilev teaches or suggests a system comprising:
at least one nontransitory processor-readable storage medium that stores at least one of processor-executable instructions or data (see Fig. 4-7); and
at least one processor communicably coupled to the at least one nontransitory processor-readable storage medium, in operation the at least one processor (see Fig. 4-7):
receives medical image data which represents an anatomical structure (see Fig. 1, 2; para. receive a medical image to be analyzed for lesions.);
processes the received image data through at least one convolutional neural network (CNN) to output predictions (see Fig. 1; para. 0002 - detect a region of interest via the trained CNN and generate a bounding box around the detected region of interest. The processor can also reduce a dimension of the region of interest based on the feature maps. The processor can also further generate a semantic description of the region of interest via the trained fully connected layers; para. 0037 - feature maps may be sent from the CNN 104 to both the region proposal network. region proposal network 106 can be trained to predict an ROI bounding box coordinates and a bounding box score; para. 0038 – predict multiple labels simultaneously. For example, the labels may be semantic descriptors.) comprising:
one or more abnormality location proposals, each including indicating a ... location of a corresponding abnormality (see para. 0002 - detect a region of interest via the trained CNN and generate a bounding box around the detected region of interest. The processor can also reduce a dimension of the region of interest based on the feature maps. The processor can also further generate a semantic description of the region of interest via the trained fully connected layers; para. 0013 - Using such features, the systems may be able to segment and characterize lesions, and, based on it, make a diagnosis; para. 0037 – the region proposal network 106 can be trained to predict an ROI bounding box coordinates and a bounding box score. region of interest candidates with associated bounding boxes and bounding box scores box score; para. 0040 – bounding boxes surrounding one or more detected lesions and a semantic description for each of the one or more detected lesions);
one or more abnormality class probabilities associated with each of the one or more abnormality location proposals (see para. 0037 – the region proposal network 106 can be trained to predict an ROI bounding box coordinates and a bounding box score. region of interest candidates with associated bounding boxes and bounding box scores box score; para. 0038 - multi-attribute net 110 may receive the region of interest candidates with bounding boxes from the RPN 106 and corresponding feature map for each region of interest from the CNN. predict multiple labels simultaneously. For example, the labels may be semantic descriptors; para. 0039 - p is the predicted probability that the ROI is in class c.);
stores the predictions in the at least one nontransitory processor-readable storage medium (see Fig. 4-7; para. 0040 - system 100 may then generate and display a diagnostic image including the medical image, the bounding box, and the semantic description; para. 0074 - various software components discussed herein may be stored on the tangible, non-transitory, computer readable medium 700, as indicated in FIG. 7.); and
presents the predictions via a user interface, wherein at least one abnormality is indicated (see Fig. 2; para. 0013 - Using such features, the systems may be able to segment and characterize lesions, and, based on it, make a diagnosis; para. 0037 – the region proposal network 106 can be trained to predict an ROI bounding box coordinates and a bounding box score. region of interest candidates with associated bounding boxes and bounding box scores box score; para. 0040 - system 100 may then generate and display a diagnostic image including the medical image, the bounding box, and the semantic description.).
Kisilev does not explicitly disclose contour indicating; changes to one or more of abnormality opacity, morphology, likelihood of malignancy, possible diagnosis or diagnoses, or the likelihood of an individual diagnosis compared to a prior exam; with at least one of a segmentation contour or a segmentation mask overlay in accordance with the contour.
Assmann teaches or suggests contour indicating; with at least one of a segmentation contour or a segmentation mask overlay in accordance with the contour (see Fig. 2-4; para. 0025 - size and exact shape of the tumor are determined automatically as a result from the magnetic resonance image data, and the details relating to the metabolism of the tumor, that is to say its composition and thus the type of the tumor, are known from the radionuclide emission tomography image data. A diagnosis or planning of operations or treatments is possible on the basis of such data; para. 0026 - magnetic resonance images are firstly segmented independently of the radionuclide emission tomography image data. A mask that is superposed in a common image on the radionuclide emission tomography image data is compiled on the basis of the contours of the target structure that are determined in the process. In conventional radionuclide emission tomography images, to be specific, the poor spatial resolution generally means that it is not individual tumors which are visible, but only a spatially unspecifiable accumulation of increased metabolism. By displaying the contours from the magnetic resonance images, it is thus possible to determine whether what is involved is an individual tumor or a number of tumors and-if the latter is the case-which part of the metabolic accumulation stems from which tumor.).
Accordingly, it would have been obvious to one having ordinary skill before the effective filing date of the claimed invention to modify the system and method, taught in Kisilev, to include contour indicating; with at least one of a segmentation contour or a segmentation mask overlay in accordance with the contour for the purpose of efficiently using masking and displaying contours to visualize tumor locations and characteristics, improving medical diagnosis and planning, as taught by Assmann (0025 and 0026).
Abedini further teaches or suggests changes to one or more of abnormality opacity, morphology, likelihood of malignancy, possible diagnosis or diagnoses, or the likelihood of an individual diagnosis compared to a prior exam (see Fig. 1-3, and 5; para. 0018 - have the images analyzed for a determination of whether skin cancer is likely, and whether and when appointments with skin cancer experts and/or further screenings should be scheduled. The embodiments of the present invention facilitate access to skin lesion monitoring systems, and provide improved methods for determining when follow-up screenings of potentially cancerous lesions should be conducted. Embodiments of the present invention enable a sufficient frequency of screening for those patients who may have limited access to health care providers; para. 0031- machine learning- based engines, such as, for example, convolutional neural network (CNN) engines CNN-1 (331), CNN-2 (332) and CNN-3 (333), which evaluate the current and past images of patient's lesions, and metadata differences when determining whether lesions are or will become cancerous; para. 0036 - processes qualitative differences 327 between images at different times, such as, for example, differences in size of lesions, shape (e.g., contour) of lesions, symmetry/asymmetry of the lesions, nature of boundaries at edges of the lesions (e.g., abrupt, gradual), and/or color of the lesion ( e.g., brown, black, yellow or gray). In accordance with a non-limiting embodiment of the present invention, the lesion analysis engine 106 is configured to detect qualitative differences 327, such as in size, shape, symmetry/ asymmetry, nature of boundaries and color which are not detectable or discernable by the human eye; para. 0037 - processes the outputs ofCNN-1 and CNN-2 to yield a combined result, which is transmitted to the prediction module 240, which outputs the probability of a lesion to be one or more of a predefined set of diseases, or none of the predefined set of diseases; para. 0054 - calculates size and pattern changes of the lesion in the last six months based on previous images stored in the database 109; para. 0055 - lesion analysis engine 106 analyzes the changes, and identifies the changes as trending toward cancer; para. 0058 - comparison can be to determine differences in features in lesion images, such as, for example, changes in size, shape, symmetry/asymmetry, nature of boundaries and colors which may not be detectable or discernable by the human eye; para. 0059 – comparing can be performed using a plurality of machine learning based engines (e.g., CNN-1, CNN-2 and CNN-3); para. 0060 - CNN can receive two consecutive images of a lesion as inputs ( e.g., at time 1 and time 2), and predict if the lesion progressed from benign to a diseased (e.g., cancerous) state or not; para. 0062 - CNN determines arithmetical differences, while another CNN determines qualitative differences between the one or more current images and the one or more previous images of the lesion.).
Accordingly, it would have been obvious to one having ordinary skill before the effective filing date of the claimed invention to modify the system and method, taught in Kisilev, to include changes to one or more of abnormality opacity, morphology, likelihood of malignancy, possible diagnosis or diagnoses, or the likelihood of an individual diagnosis compared to a prior exam for the purpose of efficiently making determinations based on differences processed by a cnn, improving diagnosis and future screening planning, as taught by Abnedini ( 0018, 0031, 0055, and 0060).
Claim(s) 90 and 93:
Claim(s) 90 and 93 correspond to Claim 54, and thus, Kisilev, Assmann, and Abedini teach or suggest the limitations of claim(s) 90 and 93 as well.
Claim 57:
Kisilev further teaches or suggests wherein the locations of the one or more abnormality location proposals are defined based on at least one of the coordinates of a regular bounding box, segmentations of the abnormalities, or one or more individual coordinates representing the location of the abnormality (see para. 0002 - detect a region of interest via the trained CNN and generate a bounding box around the detected region of interest. The processor can also reduce a dimension of the region of interest based on the feature maps. The processor can also further generate a semantic description of the region of interest via the trained fully connected layers; para. 0013 - Using such features, the systems may be able to segment and characterize lesions, and, based on it, make a diagnosis; para. 0037 – the region proposal network 106 can be trained to predict an ROI bounding box coordinates and a bounding box score. region of interest candidates with associated bounding boxes and bounding box scores box score; para. 0040 – bounding boxes surrounding one or more detected lesions and a semantic description for each of the one or more detected lesions);
Claim 65:
Kisilev teaches or suggests wherein the at least one processor utilizes at least two CNNs to determine abnormality location and classification (see Fig. 1; para. 0035 - medical image 102 can be received by a first convolutional layer of a CNN 104. The CNN 104 is communicatively connected to a region proposal network 106. For example, the region proposal network 106 may be another CNN. The region proposal network 106 can output a region of interest 108 to a multi-attribute net 110 that is communicatively coupled to both the region proposal network 106 and the CNN 104; para. 0038 - multi-attribute net 110 may receive the region of interest candidates with bounding boxes from the RPN 106 and corresponding feature map for each region of interest from the CNN. predict multiple labels simultaneously. For example, the labels may be semantic descriptors; para. 0041 – system 100 can include fewer or additional components not illustrated in FIG. 1 (e.g., additional CNNs, convolutional layers, regions of interest, descriptors, etc.).
Claim 67:
Kisilev further teaches or suggests wherein the at least one processor utilizes one CNN to determine the classification of abnormalities whose locations are already known or suspected (see para. 0015 - may receive a medical image to be analyzed for lesions. image may include one or more lesions or be an image of a location that is prone to producing malignant tumors; para. 0037 - region proposal network 108 may be any number of region of interest candidates with associated bounding boxes and bounding box scores; para. 0038 - multi-attribute net 110 may receive the region of interest candidates with bounding boxes from the RPN 106 and corresponding feature map for each region of interest from the CNN. predict multiple labels simultaneously. For example, the labels may be semantic descriptors; para. 0048 - the medical image may be an image of a body part prone to tumors or suspected of having a tumor; para. 0059 - rank the bounding box with other bounding boxes of other detected regions of interest based on a calculated probability for each detected region of interest that each region of interest is a tumor. filter out less probable regions of interest based on calculated probabilities that a plurality of regions of interest are tumors.).
Claim 68:
Kisilev further teaches or suggests wherein the at least one processor simultaneously determines the probabilities of any of the one or more classes (see para. 0015 - identify any number of lesions, such as tumors, present in an image concurrently; para. 0038 - multi-attribute net 110 can be trained to jointly predict multiple labels simultaneously. For example, the labels may be semantic descriptors; para. 0075 - For example, two blocks shown in succession may, in fact, be executed substantially concurrently.).
Claim 69:
Kisilev further teaches or suggests wherein the at least one processor utilizes the one or more CNNs to determine characteristics of a given abnormality, wherein the characteristics include at least one of: abnormality size, opacity, morphology, likelihood of malignancy, possible diagnosis or diagnoses, likelihood of any individual diagnosis, or changes to any of abnormality size, opacity, morphology, likelihood of malignancy, possible diagnosis or diagnoses, or likelihood of any individual diagnosis compared to a prior exam (see Fig. 2; para. 0035 - each region of interest 128 may have a number of descriptors indicated by boxes. The descriptors may be, for example, shape descriptors, mass descriptors, margin descriptors, etc.; para. 0039 - the shape descriptor, the possible values may be {oval, round, irregular, 0-class}; para. 0042 - respective probabilities 210 of 0.998, 0.998, 1.000, 1.000, respectively. Probabilities closer to 1.000 indicate a higher likelihood that a region of interest includes a tumor. For example, the tumor may be benign or malignant; para. 0043 - descriptions 212, 214 further contain estimated semantic descriptor values 216, 218, 220, 222 that are embedded into predefined templates; para. 0052 - processor can generate the semantic description by populating a template description with one or more semantic descriptor values; para. 0059 - the CNN to map detected features to semantic descriptors; para. 0059 - semantic description may include one or more semantic descriptors including a shape, a boundary type, a density, or any combination thereof; para. 0074 - display a diagnostic image including the medical image, the bounding box, and the semantic description.).
Claim 71:
Kisilev further teaches or suggests wherein the at least one processor determines an overall probability of an abnormality being present in a collection of one or more images from one or both of the abnormality location proposals, or abnormality characteristics associated with the abnormality location proposals (see Fig. 2; para. 0015 - able to identify any number of lesions, such as tumors, present in an image; para. 0037 - estimate a probability of the ROI bounding box including a tumor for each proposed ROI candidate; para. 0039 - p is the predicted probability that the ROI is in class c; para. 0044 - any number of regions of interest may have been identified, probabilities calculated for each identified region of interest. to be included in the diagnostic image 200 based on their respective probabilities; para. 0059 - the CNN to map detected features to semantic descriptors.).
Claim 73:
Kisilev further teaches or suggests wherein at least some of the characteristics associated with the abnormality location proposals are abnormality size, opacity, or morphology (see Fig. 2; para. 0035 - each region of interest 128 may have a number of descriptors indicated by boxes. The descriptors may be, for example, shape descriptors, mass descriptors, margin descriptors, etc.; para. 0039 - the shape descriptor, the possible values may be {oval, round, irregular, 0-class}; para. 0042 - respective probabilities 210 of 0.998, 0.998, 1.000, 1.000, respectively. Probabilities closer to 1.000 indicate a higher likelihood that a region of interest includes a tumor. For example, the tumor may be benign or malignant; para. 0043 - descriptions 212, 214 further contain estimated semantic descriptor values 216, 218, 220, 222 that are embedded into predefined templates; para. 0052 - processor can generate the semantic description by populating a template description with one or more semantic descriptor values; para. 0059 - the CNN to map detected features to semantic descriptors; para. 0059 - semantic description may include one or more semantic descriptors including a shape, a boundary type, a density, or any combination thereof; para. 0074 - display a diagnostic image including the medical image, the bounding box, and the semantic description.).
Claim 77:
Kisilev further teaches or suggests wherein the at least one CNN comprises one or more of a backbone CNN, a classification CNN, or a bounding box regression CNN (see Fig. 1; para. 0035 - medical image 102 can be received by a first convolutional layer of a CNN 104. The CNN 104 is communicatively connected to a region proposal network 106. For example, the region proposal network 106 may be another CNN. The region proposal network 106 can output a region of interest 108 to a multi-attribute net 110 that is communicatively coupled to both the region proposal network 106 and the CNN 104; para. 0038 - multi-attribute net 110 may receive the region of interest candidates with bounding boxes from the RPN 106 and corresponding feature map for each region of interest from the CNN. predict multiple labels simultaneously. For example, the labels may be semantic descriptors; para. 0041 – system 100 can include fewer or additional components not illustrated in FIG. 1 (e.g., additional CNNs, convolutional layers, regions of interest, descriptors, etc.).
Claim 78:
Kisilev further teaches or suggests wherein the at least one CNN comprises a backbone CNN that includes at least one of a classification CNN or a segmentation CNN (see Fig. 1; para. 0035 - medical image 102 can be received by a first convolutional layer of a CNN 104. The CNN 104 is communicatively connected to a region proposal network 106. For example, the region proposal network 106 may be another CNN. The region proposal network 106 can output a region of interest 108 to a multi-attribute net 110 that is communicatively coupled to both the region proposal network 106 and the CNN 104; para. 0038 - multi-attribute net 110 may receive the region of interest candidates with bounding boxes from the RPN 106 and corresponding feature map for each region of interest from the CNN. predict multiple labels simultaneously. For example, the labels may be semantic descriptors; para. 0041 – system 100 can include fewer or additional components not illustrated in FIG. 1 (e.g., additional CNNs, convolutional layers, regions of interest, descriptors, etc.).
Claim 91:
Kisilev further teaches or suggests causing a display to present one or more of the generated abnormality location proposals (see Fig. 2.).
Claim 92:
Kisilev further teaches or suggests causing the display to present only those abnormality location proposals with greater than a threshold of confidence (see Fig. 2; para. 0051 - processor can filter out less probable regions of interest based on calculated probabilities that a plurality of regions of interest are tumors; para. 0059 – semantic descriptor module 430 can filter out less probable regions of interest based on calculated probabilities that a plurality of regions of interest are tumors. The region of interest may include a higher probability than other regions of interest in the plurality of regions of interest; para. 0074 - regions with calculated probabilities below a threshold probability may be filtered out. generate and display a diagnostic image including the medical image, the bounding box, and the semantic description.).
Claim 94:
Kisilev further teaches or suggests wherein the likelihood of a given class of abnormality is visually indicated with the location proposal (see Fig. 2; para. 0043 - Probabilities closer to 1.000 indicate a higher likelihood that a region of interest includes a tumor. For example, the tumor may be benign or malignant.).
Claim 95:
Kisilev further discloses wherein the classes of abnormality include at least one of diagnoses or anatomical structures (see Fig. 2; para. 0043 - For example, the tumor may be benign or malignant. diagnostic image 200 further includes an automatically generated textual description of potential tumors within the regions of interest 202, 204, 206, 208. For example, the region of interest 202 is associated with the description "an irregular, non-homogenous mass with indistinct margins" 212. The regions of interest 204, 206, 208 are associated with the description "an oval, homogenous mass with circumscribed margins" 214. The descriptions 212, 214 further contain estimated semantic descriptor values 216, 218, 220, 222 that are embedded into predefined templates. For example, the description "an irregular, non-homogenous mass with indistinct margins" 212 contains a shape descriptor value of "irregular" 216, a density descriptor value of "non-homogenous" 218, and a margin descriptor value of "indistinct" 220. In addition, the description "an oval, homogenous mass with circumscribed margins" 214 includes a shape descriptor value of "oval" 222, a density descriptor value of "homogenous" 224, and a margin descriptor value of "circumscribed" 226; para. 0059 - the semantic description may include one or more semantic descriptors including a shape, a boundary type, a density, or any combination thereof.).
Claim(s) 72 is/are rejected under 35 U.S.C. 103 as being unpatentable over Kisilev, in view of Assmann, in view of Abedini, and further in view of Amit et al., US Publication no. 2019/0236782 (“Amit”).
Claim 72:
Kisilev does not explicitly disclose wherein at least some of the characteristics associated with the abnormality location proposals are derived from the underlying image pixel data associated with the abnormality location.
Amit teaches or suggests wherein at least some of the characteristics associated with the abnormality location proposals are derived from the underlying image pixel data associated with the abnormality location (see para. 0013 - analyzing is performed by computing a score image according to a score assigned to each pixel according to salient values above a threshold within a region around the pixel; para. 0082 - mapping between the raw image pixels and the feature vectors used for classification; para. 0118 – saliency maps may be summed (e.g., values of corresponding pixels are added together) to create a heat map; para. 0119 - heatmap may be indicative of the location of the detected lesion(s), for example, peak intensity point(s) of the heatmap associated with a multi-channel image representation classified as an indication of lesion and/or malignancy may be indicative of the location of the lesion and/or malignancy.).
Accordingly, it would have been obvious to one having ordinary skill before the effective filing date of the claimed invention to modify the system and method, taught in Kisilev, to include wherein at least some of the characteristics associated with the abnormality location proposals are derived from the underlying image pixel data associated with the abnormality location for the purpose of efficiently indicating and classifying lesion malignancy using pixel features, improving tumor detection using CNNs, as taught by Amit (0119).
Claim(s) 80 is/are rejected under 35 U.S.C. 103 as being unpatentable over Kisilev, in view of Assmann, in view of Abedini, and further in view of Tariq et al., US Publication no. 2019/0392268 (“Tariq”).
Claim 80:
Kisilev further teaches or suggests wherein the at least one CNN is trained with loss (see para. 0039 - regional proposal network 106 and the multi-attribute net 110 can be concurrently trained using the same data set during training of the CNN 104. For example, during the training of the CNN 104, a batch of positive and negative ROI candidates may be taken from two images that may be randomly chosen from a training set of images. A loss function may be defined as a multi-label loss and calculated for each batch.).
Kisilev does not explicitly disclose trained with a focal loss that corresponds to a modification of standard cross entropy loss such that the loss of predictions whose probabilities are close to the true prediction are downweighted such that their values are reduced when compared to cross entropy loss.
Tariq teaches or suggests trained with a focal loss that corresponds to a modification of standard cross entropy loss such that the loss of predictions whose probabilities are close to the true prediction are downweighted such that their values are reduced when compared to cross entropy loss (see para. 0032- training the ML model using a focal loss function. In some instances, the focal loss function may modify the loss computed by a cross entropy loss function (or any other loss function) so that the loss (errors) calculated for well classified ROIs are downweighted and the loss calculated for poorly-classified ROIs is less weighted; para. 0085 - the focal loss function downweights the error calculated for a well-classified example so that, even if there are many well-classified examples, the effect of the minimal errors each produces will have less of an effect training.).
Accordingly, it would have been obvious to one having ordinary skill before the effective filing date of the claimed invention to modify the system and method, taught in Kisilev, to include trained with a focal loss that corresponds to a modification of standard cross entropy loss such that the loss of predictions whose probabilities are close to the true prediction are downweighted such that their values are reduced when compared to cross entropy loss for the purpose of efficiently training a ML model using downweighting to reduce the effect during training, improving ML model training, as taught by Tariq (0032 and 0085).
Claim(s) 82 and 83 is/are rejected under 35 U.S.C. 103 as being unpatentable over Kisilev, in view of Assmann, in view of Abedini, and further in view of Akselrod-Ballin et al., US Publication no. 2019/0304092 (“Akselrod”).
Claim 82:
Kisilev does not explicitly disclose wherein the at least one CNN is trained using patches extracted from full size training images.
Akselrod teaches or suggests wherein the at least one CNN is trained using patches extracted from full size training images (see Fig. 1; para. 0013 - improvement in computational efficiency arises, for example, since the patch feature vectors are computed only a single time during a training round by the first stage of the deep CNN, and/or since the second stage includes a small number (e.g., 3 or other value, such as 4, or 4) of fully connected layers; para. 0116 - respective anatomical training image is decomposed into patches; para. 0124 - feature representation is computed for each patch. The feature representation may be implemented as a feature vector; para. 0129 - deep convolutional neural network is trained for detecting an indication of likelihood of abnormality for a target anatomical image according to the patches of the anatomical training images, the identified patch with highest probability value, and the probability of each respective anatomical training image; para. 0194 - Lesions in the training data set present a large scale variability of over IO scale factor. Yet the patches extracted for feeding into the deep CNN described herein are at fixed size and aim to alert for a suspicious finding, rather than exact segmentation of the lesion.).
Accordingly, it would have been obvious to one having ordinary skill before the effective filing date of the claimed invention to modify the system and method, taught in Kisilev, to include wherein the at least one CNN is trained using patches extracted from full size training images for the purpose of efficiently training a CNN using patches from training images, improving computational efficiency of the device training the CNN as well as alert effectiveness, as taught by Akselrod (0013 and 0194).
Claim 83:
Akselrod further teaches or suggests wherein inference is performed using at least one of (a) patches extracted from full size images or (b) full size images without extracting patches (see Fig. 1, 3; para. 0013 - improvement in computational efficiency arises, for example, since the patch feature vectors are computed only a single time during a training round by the first stage of the deep CNN, and/or since the second stage includes a small number (e.g., 3 or other value, such as 4, or 4) of fully connected layers; para. 0116 - respective anatomical training image is decomposed into patches; para. 0124 - feature representation is computed for each patch. The feature representation may be implemented as a feature vector; para. 0129 - deep convolutional neural network is trained for detecting an indication of likelihood of abnormality for a target anatomical image according to the patches of the anatomical training images, the identified patch with highest probability value, and the probability of each respective anatomical training image; para. 0161 - target anatomical training image is decomposed into patches; para. 0163 - probability that the respective patch includes an indication of abnormality is computed by the trained deep CNN according to the feature representation of the respective patch; para. 0164 - probability indicative of likelihood of abnormality is set for the target anatomical image; para. 0194 - Lesions in the training data set present a large-scale variability of over IO scale factor. Yet the patches extracted for feeding into the deep CNN described herein are at fixed size and aim to alert for a suspicious finding, rather than exact segmentation of the lesion.).
Accordingly, it would have been obvious to one having ordinary skill before the effective filing date of the claimed invention to modify the system and method, taught in Kisilev, to include wherein inference is performed using at least one of (a) patches extracted from full size images or (b) full size images without extracting patches for the purpose of efficiently training and using a CNN using patches from training and target images, improving computational efficiency of the device training and using the CNN as well as alert effectiveness, as taught by Akselrod (0013 and 0194).
Claim(s) 96 is/are rejected under 35 U.S.C. 103 as being unpatentable over Kisilev, in view of Assmann, in view of Abedini, and further in view of White, US Publication 2017/0325783 (“White”).
Claim 96:
Kisilev does not explicitly disclose wherein the contour includes vertices connected by non-linear spline segments.
White teaches or suggests wherein the contour includes vertices connected by non-linear spline segments (see Fig. 3C, 5A, 5B, para. 0036 - connects adjacent control points 354 with a plurality of segments 356 and in one embodiment, calculates an approximate center point 355 of the control points 354. In the illustrated embodiment of FIG. 3C, the segments 356 comprise cubic splines between adjacent control points 354. Allowing the user to draw or input the relatively few control points 354 along the boundary 352 of an anatomical structure (e.g., a heart wall) and joining the control points with smoothly connected cubic spline segments can significantly reduce the amount of time spent by the user defining the boundaries of the anatomical structure in one or more images. Moreover, the cubic splines have a curve-like shape which can be naturally very consistent with curves along the anatomical structures; para. 0040 - plurality of segments 556 (e.g., cubic splines) connect adjacent control points 554 in the individual key frames 550a and 550c. The system (e.g., the system 131 of FIG. 1) can automatically generate control points 554b for the in between frame 550b by interpolation and/or morphing along lines 559 to define the control points for the inbetween frames.).
Accordingly, it would have been obvious to one having ordinary skill before the effective filing date of the claimed invention to modify the system and method, taught in Kisilev, to include wherein the contour includes vertices connected by non-linear spline segments for the purpose of efficiently designating contours using vertices and spline curves, improving natural appearance of anatomical boundaries with a relatively low number of data points, as taught by White (0036).
Response to Arguments
Applicant’s further arguments have been considered but are not persuasive because the arguments do not correspond to the rationales as used in the current rejection.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Andrew T McIntosh whose telephone number is (571)270-7790. The examiner can normally be reached M-Th 8:00am-5:30pm.
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, Tamara Kyle can be reached at 571-272-4241. 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.
/ANDREW T MCINTOSH/Primary Examiner, Art Unit 2144
Prosecution Insights