# US 12,594,046 B2 · Method and apparatus for assisting diagnosis of cardioembolic stroke by using chest radiographic images > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,594,046 B2 - Document ID: US12594046B2 - Status: Granted - Kind: B2 - Application number: 18/028,307 - Filed: 2021-09-24 - Published: 2024-02-15 - Granted: 2026-04-07 - Earliest claimed: 2020-09-25 - Assignees: Seoul National University Hospital - Applicants: Seoul National University Hospital - Inventors: Han-Gil Jeong; Tackeun Kim - Web page: https://patentgrep.com/patent/US12594046B2 ## CPC - A61B 6/503: {for diagnosis of the heart} - A61B 6/5217: {extracting a diagnostic or physiological parameter from medical diagnostic data} - G06T 7/00: Image analysis - G06T 7/0012: {Biomedical image inspection} - G16H 30/40: for processing medical images, e.g. editing - G16H 50/20: for computer-aided diagnosis, e.g. based on medical expert systems - G06T 2207/10116: X-ray image - G06T 2207/20081: Training; Learning - G06T 2207/20084: Artificial neural networks [ANN] - A61B: DIAGNOSIS; SURGERY; IDENTIFICATION (usefulness limited to only animals A61D) - G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL - G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA ## Abstract Embodiments relate to a method for assisting the diagnosis of cardioembolic stroke, in which the method is performed by a processor and is to assist the diagnosis of cardioembolic stroke by using chest radiographic images, the method comprising the steps of: acquiring chest radiographic images to be classified of a subject; and acquiring cardioembolic stroke diagnosis auxiliary information from the chest radiographic images to be classified of the subject, by using a neural network model, wherein the neural network model is configured to extract features from the chest radiographic images to be classified of the subject, and determine, based on the extracted features, whether the subject belongs to a group indicative of having cardioembolic stroke. ## Claims ### Claim 1 A method for assisting diagnosis of cardioembolic stroke, in which the method is performed by a processor and is to assist diagnosis of cardioembolic stroke by using chest radiographic images, comprising: acquiring chest radiographic images to be classified of a subject; and generating cardioembolic stroke auxiliary information from the chest radiographic images to be classified of the subject by using a neural network model; wherein the neural network model is configured to extract features from the chest radiographic images to be classified of the subject, and determine, based on the extracted features, whether the subject belongs to a group indicative of having cardioembolic stroke, wherein the neural network model is trained, using a plurality of training samples, to classify chest radiographic images into a cardioembolic stroke group and a non-cardioembolic stroke group, and wherein each of the plurality of training samples includes a chest radiographic image for training and label information thereof. ### Claim 2 (depends on 1) The method for assisting diagnosis of cardioembolic stroke of claim 1, wherein the neural network model has a convolutional neural network (CNN) structure. ### Claim 3 (depends on 2) The method for assisting diagnosis of cardioembolic stroke of claim 2, wherein the neural network model is learned by adjusting parameters, wherein the parameters include parameters indicating one or more of left atrial volume index, early diastolic bicuspid annular tissue velocity (E/e′), frequency of bicuspid and tricuspid valve disease, and left ventricular ejection fraction, and wherein the left atrial volume index, the E/e′, and the left ventricular ejection fraction are correlated with structural abnormalities of a left atrium that are indicative of cardioembolic stroke. ### Claim 4 (depends on 3) The method for assisting diagnosis of cardioembolic stroke of claim 3, wherein the neural network model includes a fully connected layer, wherein the fully connected layer includes a single output sigmoid layer so that the neural network model learns classification into a cardioembolic stroke group and a non-cardioembolic stroke group. ### Claim 5 (depends on 1) The method for assisting diagnosis of cardioembolic stroke of claim 1, further comprising outputting class activation map (CAM) CAM images through a gradient class activation map (Grad-CAM) model based on at least one of the acquired chest radiographic images to be classified and the information generated from the neural network model in order to identify the region of the chest radiographic images that affects diagnosis performed by the neural network model. ### Claim 6 (depends on 5) The method for assisting diagnosis of cardioembolic stroke of claim 5, wherein a suspected region including the region where the left atrium is located is indicated, by the Grad-CAM, in chest radiographic images to be classified that belong to a group indicative of having cardioembolic stroke, and wherein the region where the left atrium is located is identified based on structural features of left atrial enlargement that are characteristic of cardioembolic stroke. ### Claim 7 (depends on 1) A non-transitory computer-readable recording medium readable by a computer and storing program instructions operable by the computer, which causes a processor of the computer to perform the method for assisting diagnosis of cardioembolic stroke according to claim 1 when the program instructions are executed by the processor. ### Claim 8 An apparatus for assisting diagnosis of cardioembolic stroke, the apparatus comprising a processor and memory and being configured to perform operations comprising: acquiring chest radiographic images to be classified of a subject; and generating, using a neural network model, auxiliary information for diagnosis of cardioembolic stroke from the chest radiographic image to be classified of the subject; wherein the neural network model is configured to extract features from the chest radiographic images to be classified of the subject, and determine, based on the extracted features, whether the subject belongs to a group indicative of having cardioembolic stroke, wherein the neural network model is trained, using a plurality of training samples, to classify chest radiographic images into a cardioembolic stroke group and a non-cardioembolic stroke group, and wherein each of the plurality of training samples includes a chest radiographic image for training and label information thereof. ### Claim 9 (depends on 8) The apparatus for assisting diagnosis of cardioembolic stroke of claim 8, wherein the neural network model has a convolutional neural network (CNN) structure. ### Claim 10 (depends on 9) The apparatus for assisting diagnosis of cardioembolic stroke of claim 9, wherein the neural network model is learned by adjusting parameters, wherein the parameters include parameters indicating one or more of left atrial volume index, early diastolic bicuspid annular tissue velocity (E/e′), frequency of bicuspid and tricuspid valve disease, and left ventricular ejection fraction, and wherein the left atrial volume index, the E/e′, and the left ventricular ejection fraction are correlated with structural abnormalities of a left atrium that are indicative of cardioembolic stroke. ### Claim 11 (depends on 10) The apparatus for assisting diagnosis of cardioembolic stroke of claim 10, wherein the neural network model includes a fully connected layer, wherein the fully connected layer includes a single output sigmoid layer so that the neural network model learns classification into a cardioembolic stroke group and a non-cardioembolic stroke group. ### Claim 12 (depends on 8) The apparatus for assisting diagnosis of cardioembolic stroke of claim 8, wherein the operations further comprise outputting class activation map (CAM) images through a gradient class activation map (Grad-CAM) model based on at least one of the acquired chest radiographic images to be classified and the information generated from the neural network model in order to identify the region of the chest radiographic images that affects diagnosis performed by the neural network model. ### Claim 13 (depends on 12) The apparatus for assisting diagnosis of cardioembolic stroke of claim 12, wherein a suspected region including the region where the left atrium is located is indicated, by the Grad-CAM, in chest radiographic images to be classified that belong to a group indicative of having cardioembolic stroke, and wherein the region where the left atrium is located is identified based on structural features of left atrial enlargement that are characteristic of cardioembolic stroke.