US-12626418-B2 - Device and method for supporting biometric image finding/diagnosis

US12626418B2US 12626418 B2US12626418 B2US 12626418B2US-12626418-B2

Abstract

Provided are a device, a method, and a system for supporting biometric image finding/diagnosis, the device comprising: a processor; and a memory including one or more instructions implemented to be executed by the processor, wherein the processor; extracts a first attribute information from a first biometric image of an object on the basis of a machine learning model; changes the first attribute information of the first biometric imaged by mapping adversarial noise to the first biometric image, so as to generate a second biometric image having second attribute information; and displays the first biometric image having the first attribute information and the second biometric image having the second attribute information on a display unit.

Inventors

Sang Min Park

Assignees

XAIMED CO., LTD

Dates

Publication Date: 20260512
Application Date: 20211220
Priority Date: 20210125

Claims (13)

1 . An apparatus for supporting reading of a biometric image of a subject, comprising: a processor; and a memory including one or more sequences of instructions which, when executed by the processor, causes steps to be performed comprising: extracting a first feature information from a first biometric image of the subject based on a machine learning model; generating a second biometric image having a second feature information by mapping an adversarial noise to the first biometric image so that the first feature information of the first biometric image is changed; generating a third biometric image in which the adversarial noise is visualized, by filtering the second biometric image mapped by the adversarial noise therein; and displaying the first biometric image having the first feature information, the second biometric image having the second feature information and the third biometric image on a display device.
2 . The apparatus of claim 1 , wherein the steps further comprises: generating images in which a specific area of each of the first biometric image and the second biometric image is enlarged; and displaying the images on the display device.
3 . The apparatus of claim 1 , wherein the steps further comprises: generating a pre-biometric image by pre-processing the first biometric image to enlarge or partially partition a specific area of the first biometric image.
4 . The apparatus of claim 1 , wherein the adversarial noise comprises at least one of gradation levels of R, G, and B pixels of the first biometric image, a color of R, G, and B pixels of the first biometric image, and a contrast ratio of the first biometric image.
5 . The apparatus of claim 1 , wherein the second biometric image is generated by repeatedly mapping by the adversarial noise to the first biometric image at least one or more times so that a prediction value of the first feature information obtained based on the machine learning model for the first biometric image converges to a set value.
6 . The apparatus of claim 5 , wherein the number of second biometric image is dependent on the number of the set value.
7 . The apparatus of claim 1 , wherein the first feature information is extracted by utilizing a clinical information of the subject.
8 . The apparatus of claim 1 , wherein the first biometric image includes a fundus image, a first feature information of the fundus image comprises at least one of a cup-to-disk ratio, a thickness change for a disc rim thinning, a contrast for a retinal nerve fiber layer defect and, a location of a retinal hemorrhage included in the fundus image.
9 . The apparatus of claim 1 , wherein the adversarial noise is mapped into the first biometric image in an unit of grouped cells, which spatially group image pixels composing the first biometric image.
10 . The apparatus of claim 9 , wherein the image pixels are grouped based on a color of an adjacent image pixels each other.
11 . A method for supporting reading of a biometric image of a subject, comprising: extracting a first feature information from a first biometric image of the subject based on a machine learning model; generating a second biometric image having a second feature information by mapping an adversarial noise to the first biometric image so that the first feature information of the first biometric image is changed; generating a third biometric image in which the adversarial noise is visualized, by filtering the second biometric image mapped by the adversarial noise therein; and displaying the first biometric image having the first feature information, the second biometric image having the second feature information and the third biometric image on a display device.
12 . The method of claim 11 , further comprising: generating images in which a specific area of each of the first biometric image and the second biometric image is enlarged; and displaying the images on the display device.
13 . The method of claim 11 , further comprising: generating a pre-biometric image by pre-processing the first biometric image so that a specific area of the first biometric image is enlarged or partially partitioned.

Description

TECHNICAL FIELD The present disclosure relates to diagnosing disease of biometric images, more particularly, to an apparatus and method for supporting reading of biometric images using a machine learning model. BACKGROUND ART With development of artificial intelligence learning models, many machine learning models are being used to read medical images. For example, the machine learning models such as Convolutional Neural Networks (CNN), Deep Neural Networks (DNN), Recurrent Neural Networks (RNN), and Deep Belief Networks (DBN) are being applied to detect, classify, and characterize the medical images. The machine learning models are currently being used to support an image reading, an image finding, an image diagnosis to predict a disease of a patient. More specifically, a method of supporting the image reading, the image finding, and the image diagnosis of the medical image is to obtain the biometric image from the patient, extract feature from the fundus image based on the machined learning models, provide the feature to a practitioner, and predict the patient's disease based on it. In this case, the feature includes various information for the medical image. However, even if the feature of the medical image is extracted based on the machine learning model, if the learning information input to the machine learning model is inadequate or insufficient for various factors such as a lack of learning data input to the machine learning mode, differences in environments of medical imaging (e.g., a health check-up center, a private ophthalmology clinic, a general hospital, etc.), a difference of learning data (e.g., only a fundus image of a normal person, only a fundus image of an abnormal person), a difference of an imaging apparatuses, an entity such as a medical practitioner can receive incorrect information from the machined learning model. For example, differences in learning information are the lack of learning data input to the learning model, differences in imaging environments (e.g., health examination centers, private ophthalmology hospitals, general ophthalmology hospitals), and groups (e.g., only normal people, only abnormalities). It may be a difference between a person, a normal person and an abnormal person), a difference between an imaging device, and the like. These various factors can lead to erroneous prediction of the patient's disease. Thus, even though the learning information is poor, there is a need for systems and methods that can do the more accurate prediction of the patient's disease using the learning information image and can explain why such the patient's disease was predicted. DESCRIPTION OF EMBODIMENTS Technical Problem An embodiment of the present disclosure is to provide an apparatus for supporting reading of a biometric image capable of improving the accuracy and reliability of biometric image reading based on a machine learning model. Another embodiment of the present disclosure is to provide an apparatus for supporting reading of a biometric image capable of explaining the reason for biometric image reading. Solution to Problem In one aspect of the present disclosure, an apparatus for supporting reading of a biometric of a subject includes a processor and a memory including one or more sequences of instructions which, when executed by the processor, causes steps to be performed comprising: extracting a first feature information from a first biometric image of the subject based on a machine learning model; generating a second fundus image having a second feature information by mapping an adversarial noise to the first biometric image so that the first feature information of the first biometric image is changed; and displaying the first biometric image having the first feature information and the second biometric image having the second feature information on a display device. Desirably, the steps further may include generating images in which a specific area of each of the first biometric image and the second biometric image is enlarged; and displaying the images on the display device. Desirably, the steps further may include generating a third biometric image in which an adversarial noise is visualized, by filtering the second biometric image mapped by the adversarial noise therein; and displaying the third biometric image on the display device. Desirably, the steps further may include generating a pre-biometric image by pre-processing the first biometric image to enlarge or partially partition a specific area of the first biometric image. In another aspect of the present disclosure, a method for supporting reading of a biometric image of a subject includes extracting a first feature information from a first biometric image of the subject based on a machine learning model; generating a second biometric image having a second feature information by mapping an adversarial noise to the first biometric image so that the first feature information of the first biometric image is cha