CN-122023268-A - Method and device for detecting hypertensive retinopathy

CN122023268ACN 122023268 ACN122023268 ACN 122023268ACN-122023268-A

Abstract

The invention provides a method and a device for detecting hypertensive retinopathy, wherein the method comprises the steps of detecting hypertensive retinopathy characteristics of a fundus image to be detected, determining whether the first hypertensive retinopathy characteristics or the second hypertensive retinopathy characteristics exist in the fundus image to be detected if the hypertensive retinopathy characteristics represent the first hypertensive retinopathy characteristics or the second hypertensive retinopathy characteristics of the fundus image to be detected, indicating arteriovenous vascular intersection characteristics if the hypertensive retinopathy characteristics represent the first hypertensive retinopathy characteristics and the second hypertensive retinopathy characteristics not exist in the fundus image to be detected, determining vascular parameters of arteriovenous trunk vessels in the fundus image to be detected, predicting the vascular parameters by using a classification model, and outputting a hypertensive retinopathy classification result. Thus, the evaluation of hypertensive retinopathy is made more efficient.

Inventors

DONG ZHOU
Ling Saiguang
KE XIN

Assignees

依未科技(北京)有限公司

Dates

Publication Date: 20260512
Application Date: 20251231

Claims (10)

1. A method for detecting hypertensive retinopathy is characterized in that, Detecting the characteristics of the hypertensive retinopathy of the fundus image to be detected; If the detection result of the feature of the hypertensive retinopathy indicates that the first feature of the hypertensive retinopathy or the second feature of the hypertensive retinopathy exists in the fundus image to be detected, determining that the hypertensive retinopathy exists in the fundus image to be detected, wherein the first feature of the hypertensive retinopathy is used for indicating the characteristic of the optic disc edema or simultaneously indicating the bleeding and the soft seepage of the appointed shape, and the second feature of the hypertensive retinopathy is used for indicating the characteristic of the arteriovenous vascular intersection; If the detection result of the feature of the hypertensive retinopathy indicates that the first feature of the hypertensive retinopathy and the second feature of the hypertensive retinopathy do not exist in the fundus image to be detected, determining the blood vessel parameters of arterial and venous trunk blood vessels in the fundus image to be detected; and predicting the blood vessel parameters by using a classification model, and outputting a classification result of the hypertensive retinopathy.
2. The method of claim 1, wherein the step of determining the position of the substrate comprises, The method comprises the steps of detecting first hypertension retinopathy characteristics of a fundus image to be detected, determining that hypertension retinopathy exists in the fundus image to be detected if the detection result indicates that the first hypertension retinopathy characteristics exist in the fundus image to be detected, and performing arteriovenous vessel segmentation processing on the fundus image to be detected if the detection result indicates that the first hypertension retinopathy characteristics do not exist in the fundus image to be detected, so as to generate arteriovenous segmentation results; And if the detection result represents that the arteriovenous segmentation result has the second hypertensive retinopathy characteristic, determining that the hypertensive retinopathy exists in the fundus image to be detected.
3. A method according to claim 1 or 2, characterized in that, Identifying the characteristic of optic disc edema in the fundus image to be detected; If the identification result represents that the characteristic of the optic disc edema exists in the fundus image to be detected, determining that the hypertensive retinopathy exists in the fundus image to be detected; If the identification result indicates that the characteristic of optic disc edema does not exist in the fundus image to be detected, extracting bleeding points and soft seepage from the fundus image to be detected; And if the bleeding point shape is strip-shaped, determining that the hypertensive retinopathy exists in the fundus image to be detected.
4. The method according to claim 2, wherein the performing an arteriovenous vessel segmentation process on the fundus image to be measured to generate an arteriovenous segmentation result comprises: Performing image preprocessing on the fundus image to be detected to generate a fundus blood vessel image; Performing example segmentation on the fundus blood vessel image to obtain a blood vessel example segmentation result, wherein the blood vessel example segmentation result is used for indicating a blood vessel segmentation result with blood vessel attributes output from a video disc boundary; Calibrating the blood vessel attribute of each blood vessel in the blood vessel example segmentation result, and outputting a blood vessel example segmentation result after calibration; numbering all blood vessels in the blood vessel example segmentation result based on a preset rule, and outputting a blood vessel marking result; And identifying an arteriovenous vessel based on the vessel marking result to obtain an arteriovenous segmentation result.
5. The method according to claim 2, wherein the performing the second hypertensive retinopathy feature detection on the arteriovenous segmentation result, if the detection result indicates that the arteriovenous segmentation result has the second hypertensive retinopathy feature, determining that there is hypertensive retinopathy in the fundus image to be detected, includes: Extracting the crossing position of the artery and vein segmentation result; determining whether a cross point feature exists at the cross point location; and if the intersection point features exist at the intersection position, determining that the hypertensive retinopathy exists in the fundus image to be detected.
6. The method according to claim 1, wherein determining the vascular parameters of the arterial and venous main blood vessels in the fundus image to be measured comprises: Based on the arteriovenous segmentation result, the arteriovenous main blood vessels at the two ends of the video disc are identified to obtain two groups of arteriovenous main blood vessels, wherein the arteriovenous main blood vessels are used for indicating the arterial main blood vessels and the venous main blood vessels which are positioned at any end of the video disc and have adjacent relation; For any group of arterial and venous trunk blood vessels, respectively calculating a first arterial blood vessel parameter of an arterial trunk blood vessel and a first venous blood vessel parameter of a venous trunk blood vessel at a preset distance from a central point of a video disc, determining an arterial and venous parameter ratio based on the first arterial blood vessel parameter and the first venous blood vessel parameter, and obtaining a plurality of arterial and venous parameter ratios corresponding to different preset distances based on the arterial and venous parameter ratio corresponding to each preset distance, wherein the arterial and venous parameter ratio comprises an arterial and venous vessel diameter ratio and an arterial and venous average brightness ratio; for any group of arterial and venous trunk blood vessels, respectively calculating the average curvature corresponding to the arterial trunk blood vessel and the average curvature corresponding to the venous trunk blood vessel at the preset distance from the center point of the optic disc, determining the average curvature corresponding to the arterial trunk blood vessel and the average curvature corresponding to the venous trunk blood vessel as the arterial and venous average curvature corresponding to the preset distance; and taking the ratio of the arteriovenous parameters and the average tortuosity of the arteriovenous parameters as the blood vessel parameters of the arteriovenous main blood vessel in the fundus image to be measured.
7. The method of claim 6, wherein the classification model is obtained by: Acquiring a plurality of target fundus images; Performing image preprocessing on the target fundus image to generate a target fundus blood vessel image aiming at any one of the target fundus images; an arteriovenous blood vessel segmentation process is carried out on the target fundus blood vessel image to generate an arteriovenous segmentation result, and an arteriovenous vessel diameter ratio, an arteriovenous average brightness ratio and an arteriovenous average curvature are determined based on the arteriovenous segmentation result; If one of the arteriovenous vessel diameter ratio, the arteriovenous average brightness ratio and the arteriovenous average curvature meets a preset condition, determining that the target fundus image has hypertensive retinopathy, and determining the arteriovenous vessel diameter ratio, the arteriovenous average brightness ratio and the arteriovenous average curvature corresponding to the target fundus image as a first training sample; if all three of the arteriovenous vessel diameter ratio, the arteriovenous average brightness ratio and the arteriovenous average curvature do not meet preset conditions, determining that the target fundus image does not have hypertensive retinopathy, and determining the arteriovenous vessel diameter ratio, the arteriovenous average brightness ratio and the arteriovenous average curvature corresponding to the target fundus image as a second training sample; And performing machine learning based on a first training sample or a second training sample corresponding to each target fundus image in the plurality of target fundus images, constructing a corresponding function model, and generating a classification model.
8. A device for detecting hypertensive retinopathy is characterized in that, The feature detection module is used for detecting the feature of the hypertensive retinopathy of the fundus image to be detected; The first determining module is used for determining that the first hypertensive retinopathy characteristic or the second hypertensive retinopathy characteristic exists in the fundus image to be detected if the first hypertensive retinopathy characteristic or the second hypertensive retinopathy characteristic exists in the fundus image to be detected according to the detection result of the hypertensive retinopathy characteristic, wherein the first hypertensive retinopathy characteristic is used for indicating the characteristic of optic disc edema or simultaneously indicating bleeding and soft permeation in a designated shape, and the second hypertensive retinopathy characteristic is used for indicating the characteristic of arteriovenous vascular intersection; the second determining module is used for determining the vascular parameters of the arterial and venous main blood vessels in the fundus image to be detected if the detection result of the hypertensive retinopathy features represents that the first hypertensive retinopathy features and the second hypertensive retinopathy features do not exist in the fundus image to be detected; And the prediction module is used for predicting the blood vessel parameters by using the classification model and outputting a classification result of the hypertensive retinopathy.
9. An electronic device comprising a processor, a memory for storing instructions executable by the processor, the processor for reading the executable instructions from the memory and executing the instructions to implement the method of any of claims 1-7.
10. A computer readable medium having stored thereon a computer program which, when executed by a processor, implements the method of any of claims 1-7.

Description

Method and device for detecting hypertensive retinopathy Technical Field The invention belongs to the technical field of image processing, and particularly relates to a method and a device for detecting hypertensive retinopathy. Background The prior art is based on the fact that the diagnosis of the hypertensive retinopathy by fundus images is carried out by visual reading of doctors. However, the naked eye can only diagnose whether the patient has hypertension based on fundus images, and there is no way to effectively grade the hypertensive retinopathy. Different ocular fundus abnormalities reflect different courses of hypertension. Traditional Keith-Wagener-Barker hypertensive retinopathy grading criteria include grade I retinal arterioles light/medium constriction with arteriole ratio > =1:2, grade ii retinal arterioles medium, severe constriction (local or diffuse), with arteriole-venous ratio <1:2 or arteriole cross compression, grade III retinal soft exudation or flame-like hemorrhage, grade IV binocular optic disc oedema. Disclosure of Invention Aiming at the problems existing in the prior art, the embodiment of the invention provides a method and a device for detecting hypertensive retinopathy, which are based on image processing technologies such as computer vision, deep learning and the like, and can automatically and quickly analyze hypertensive retinopathy in fundus images, thereby greatly saving the time of reading by doctors, enabling evaluation to be more efficient, avoiding subjective differences among doctors, realizing consistency evaluation of fundus images, and improving accuracy and reliability. According to a first aspect of the embodiment of the invention, a method for detecting hypertensive retinopathy is provided, wherein hypertensive retinopathy feature detection is carried out on a fundus image to be detected, if a hypertensive retinopathy feature detection result represents that a first hypertensive retinopathy feature or a second hypertensive retinopathy feature exists in the fundus image to be detected, the hypertensive retinopathy is determined to exist in the fundus image to be detected, wherein the first hypertensive retinopathy feature is used for indicating a disc edema feature or simultaneously generating specified shape hemorrhage and soft osmosis feature, the second hypertensive retinopathy feature is used for indicating arteriovenous vascular intersection feature, and if the hypertensive retinopathy feature detection result represents that the first hypertensive retinopathy feature and the second hypertensive retinopathy feature do not exist in the fundus image to be detected, vascular parameters of arteriovenous trunk blood vessels in the fundus image to be detected are determined, and a classification model is used for predicting the vascular parameters to output a hypertensive retinopathy classification result. Optionally, performing first hypertensive retinopathy feature detection on the fundus image to be detected, determining that hypertensive retinopathy exists in the fundus image to be detected if the detection result indicates that the first hypertensive retinopathy feature exists in the fundus image to be detected, performing arteriovenous vascular segmentation processing on the fundus image to be detected to generate arteriovenous segmentation results, performing second hypertensive retinopathy feature detection on the arteriovenous segmentation results, and determining that hypertensive retinopathy exists in the fundus image to be detected if the detection result indicates that the second hypertensive retinopathy feature exists in the arteriovenous segmentation results. Optionally, identifying characteristics of optic disc edema in the fundus image to be tested, determining that hypertensive retinopathy exists in the fundus image to be tested if the identification result indicates that the characteristics of optic disc edema exist in the fundus image to be tested, extracting bleeding points and soft seepage from the fundus image to be tested if the identification result indicates that the characteristics of optic disc edema do not exist in the fundus image to be tested, determining shapes of the bleeding points if the extraction result indicates that the bleeding points and the soft seepage exist in the fundus image to be tested at the same time, and determining that the hypertensive retinopathy exists in the fundus image to be tested if the shapes of the bleeding points are strip-shaped. Optionally, performing arteriovenous blood vessel segmentation processing on the fundus image to be detected to generate an arteriovenous segmentation result, wherein the arteriovenous blood vessel segmentation processing comprises the steps of performing image preprocessing on the fundus image to be detected to generate a fundus blood vessel image, performing example segmentation on the fundus blood vessel image to obtain a blood vessel example segmentation result, wherei