AS-OCT image room angle classification method based on convolutional recurrent neural network

Abstract

The invention discloses an AS-OCT image room angle classification method based on a convolutional recurrent neural network, and the method comprises the steps: carrying out the alignment operation of AS-OCT slices through a global scanning alignment method, and solving the problem of image dislocation caused by the possibility that the eyes do not move autonomously and the optical axes of the eyes are not properly placed; segmenting the iris by adopting a deep learning segmentation algorithm so as to determine an ACA region by the iris root; modeling two-dimensional images and image sequence information at the same time based on a convolutional recurrent neural network, and the classification performance of the network for narrow angles and adhesion is improved. According to the method, open, narrow and adhesive glaucoma can be accurately classified, and the world advanced level is achieved.

Description

technical field [0001] The invention relates to a classification method of AS-OCT images based on deep learning, which belongs to the field of image processing. It can not only judge open-angle and closed-angle glaucoma from a single AS-OCT image, but also can judge from AS-OCT image sequences. Identify dynamic features to enable accurate classification of open, narrow, and cohesive ACAs. Background technique [0002] Glaucoma, cataract, and diabetic retinopathy are collectively known as the "three major killers" of blindness. According to statistics, glaucoma is the second leading cause of blindness worldwide, second only to cataract in blinding rate. Not only that, glaucoma is also the world's first irreversible eye disease, and 70% of glaucoma patients will eventually lose their vision and become blind. Glaucoma is irreversible because glaucoma disease causes a series of damage to the optic nerve, resulting in gradual loss of vision or complete loss of eye disease. Gla...

AI Technical Summary

Problems solved by technology

This is because it may be difficult to distinguish between narrow-angle and fully-closed-angle regions based on only one static AS-OCT image

Images