Fundus image quality evaluation method based on blood vessel segmentation and background separation

Abstract

The invention discloses a fundus image quality evaluation method based on blood vessel segmentation and background separation, and the method comprises the following steps: 1), carrying out the bloodvessel segmentation of an input image through a pre-trained U-Net model on a DRIVE public fundus image data set; 2) multiplying the blood vessel feature map obtained in the step 1) by an original image element by element to obtain an image only containing blood vessels and background information; 3) respectively inputting the extracted feature images into convolutional neural network branches fortraining to obtain model parameters; and 4) performing quality evaluation on the test picture by using the trained convolutional neural network model. According to the invention, higher evaluation accuracy is realized, the reexamination rate of doctors is reduced, and possible treatment opportunity delay caused by repeated examination is avoided. The model provided by the invention has universality and can be embedded into various advanced convolutional neural network structures, the network performance is improved, and meanwhile, a method for fusing vascular prior knowledge and neural networkend-to-end feature extraction is provided.

Description

technical field [0001] The invention relates to the fields of medical image processing and computer vision, in particular to an image quality evaluation method based on a convolutional neural network. Background technique [0002] The fundus image is captured by a special fundus camera, which contains important physiological structures such as the optic disc, macula, and blood vessels in the retina, and is an important type of image in medical imaging. In the normal fundus image, the optic disc appears as a nearly circular bright color area, which has the strongest contrast with the background area, and is the starting area of ​​the optic nerve and blood vessels; because the macula is rich in lutein, it is in the normal fundus image It is manifested as a dark area, and the dark area has no blood vessel structure, and there is an inwardly sunken area called the fovea in the center of the macula; the blood vessels start from the optic disc area and extend to the entire interio...

AI Technical Summary

Problems solved by technology

A study involving an extensive database of retinal images reveals that more than 25 percent of images appear to be of insufficient quality for proper medical diagnosis

In addition to the financial investment required to reacquire poor quality photographs, it is inconvenient for the patient to return to the medical center for repeat fundus photographs

What's more serious is that misdiagnosis caused by poor image quality is not easy to be found, delaying the timing of treatment

Images