Fundus illumination multi-disease detection system based on regional feature set neural network

Abstract

The invention discloses a fundus illumination multi-disease detection system based on a regional feature set neural network. The system comprises a computer memory, a computer processor and a computerprogram which is stored in the computer memory and can be executed on the computer processor, wherein a trained multi-fundus disease detection network model is stored in the computer memory, and themulti-fundus disease detection network model comprises a feature extraction network, a semantic segmentation sub-network and a plurality of classifiers; when the computer processor executes the computer program, the following steps are realized: obtaining a to-be-detected original fundus picture, inputting the to-be-detected original fundus picture into the multi-fundus-disease detection network model, obtaining the probability of each type of lesion, and realizing fundus lesion identification. According to the invention, a single model can be used for detecting various fundus diseases, and the detection precision is high.

Description

technical field [0001] The invention belongs to the field of ophthalmology imaging, and in particular relates to a multi-disease detection system for fundus photographs based on a regional feature set neural network. Background technique [0002] With the improvement of medical technology level, eye health has been paid more and more attention by people. Many fundus diseases can cause irreversible damage to vision. Fundus camera imaging is currently one of the most common and easiest-to-operate ophthalmic clinical imaging methods. At present, the analysis and research of fundus photos mostly use machine learning methods, including supervised learning and unsupervised learning methods. Due to the expert annotation supervision model training, the effect and generalization of supervised learning are significantly higher than that of unsupervised learning. Among them, the method based on convolutional neural network has achieved almost unsurpassable accuracy in fundus map anal...

AI Technical Summary

Problems solved by technology

For example, the lesions of macular degeneration are concentrated in the macular area, but the macular area only accounts for about 1 / 10 of the entire fundus photo, and redundant features will interfere with the training of the model

In order to prevent redundant feature interference, the recognition model is generally trained separately for different diseases, that is, a binary classification model is trained for each disease, and multiple feature extractions are performed on a fundus map, resulting in repeated calculations, which consume a lot of time and computing power.

There is still a lot of room for improvement in the identification of multiple diseases with a single model

Images