Fundus image optic cup and optic disk segmentation method

Abstract

A fundus image optic cup and optic disk segmentation method comprises the following steps: a training stage: training a positioning network and a segmentation network, the segmentation network comprising a generation network and a discrimination network, the positioning network being independently trained, and the segmentation network training being trained by alternate iteration of the generationnetwork and the discrimination network; a prediction stage: obtaining an optic disk segmentation map through the positioning network; and performing region-of-interest cutting according to the opticdisc segmentation image, performing alignment transformation on the region-of-interest, obtaining an optic cup optic disc segmentation result image and an optic cup optic disc boundary segmentation result image through a generation network in the segmentation network, and restoring the segmentation result image to an original cutting position and size to obtain a final result image. Data with different sources is trained through iteration of the generation network and the discrimination network, and the fundus image optic cup and optic disk segmentation method can adapt to multi-domain and end-to-end segmentation optic disks, and effectively solves the problem of data difference.

Description

technical field [0001] The invention relates to the field of fundus image processing, in particular to a method for segmenting an optic cup and optic disc of a fundus picture. Background technique [0002] Eyes are an important organ for human visual perception. External light passes through human eye nerve cells to form visual signals and transmits them to the brain. The eyeball mainly includes the sclera (white of the eye), iris, cornea, lens, choroid, retina, optic nerve, pupil, and other parts [1]. Fundus image-assisted diagnosis has become the first choice for large-scale retinal disease screening due to its advantages of simple operation, low cost, and little harm to the human body. The fundus retina is an important part of vision. Ophthalmologists often use non-invasive and convenient fundus images to check the fundus retina for the diagnosis of eye diseases. The fundus image is mainly composed of blood vessels, optic disc, macula and other parts, in which there may...

AI Technical Summary

Problems solved by technology

[0004] At present, due to differences in camera models and angles, there are differences in fundus images. The above-mentioned fundus image data from different sources cannot use a unified model and parameters to segment the optic cup and optic disc.

Images