Retinal blood vessel segmentation method, device, electronic device and storage medium

By iteratively processing and weight matrix adjustment of the retinal image, adaptively imparting different loss function punishment weights to different blood vessel categories, solving the problem of poor segmentation of small blood vessels in the existing technology and achieving higher segmentation accuracy.

CN114897911BActive Publication Date: 2025-03-28WUYI UNIV
1 Cites 0 Cited by

Patent Information

Application Number
CN202210380303.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-04-12
Publication Date
2025-03-28
Estimated Expiration
2042-04-12

AI Technical Summary

Technical Problem

The prior art is difficult to effectively identify and segment tiny blood vessels in retinal vascular segmentation, resulting in poor segmentation effect.

Method used

By acquiring the retinal image and its label map, iteratively traverse each pixel point, obtaining label information, and parameterize the original weight matrix according to the label information and the preset adjustment function to generate the first corrected weight matrix. The deep learning network model is then trained based on the weight matrix and the cross entropy loss function, and adaptively imparts different penal weights to the background, coarse blood vessels and thin blood vessels.

Benefits of technology

The effect of retinal blood vessel segmentation has been improved, segmentation accuracy has been improved, especially in the segmentation of tiny blood vessels.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The embodiments of the present invention provide a retinal blood vessel segmentation method, device, electronic device and storage medium. The method includes: obtaining a retinal image and a label map corresponding to the retinal image; obtaining an original weight matrix according to the label map; iteratively traversing each pixel in the label map to obtain label information; adjusting the parameters of the original weight matrix according to the label information and a preset adjustment function to obtain a first corrected weight matrix; training a preset deep learning network model based on the first corrected weight matrix, the retinal image, the label map and a preset cross entropy loss function; and inputting the retinal image into the trained deep learning network model to obtain a retinal blood vessel segmentation result. According to the solution of the embodiments of the present invention, the effect of retinal blood vessel segmentation can be improved, the segmentation accuracy can be improved, and the segmentation of small blood vessels can be more perfect.
Need to check novelty before this filing date? Find Prior Art