Diabetic retinopathy lesion image recognition method based on attention model

Abstract

The invention provides a diabetic retinopathy lesion image recognition method based on an attention model, and the method is characterized in that the method comprises the steps: firstly, carrying out the preprocessing suitable for a medical image for an original fundus image in a data set; secondly, constructing an attention network model based on semantic segmentation to perform feature extraction; then, constructing a lesion sensing module, a feature keeping module, a feature fusion module and a head attention module, extracting lesion related information from the extracted features, and generating a lesion detection probability graph; and finally, guiding eye disease screening by using the focus detection probability graph obtained by the attention model according to the input fundus photo, and obtaining a fundus photo recognition result with diabetic retinopathy focus information. According to the method, related focus information in the fundus photo can be effectively obtained, the interpretable focus detection probability graph is generated while the diabetic retinopathy is automatically screened, ophthalmologists can be well assisted in diagnosis, and the method has wide clinical application prospects.

Description

technical field [0001] The present invention relates to the multidisciplinary technical field of computer vision, digital image processing and ophthalmology clinical medicine, specifically, relates to a kind of diabetic retinopathy lesion image recognition method based on attention model. Background technique [0002] Diabetic retinopathy is a retinal complication caused by diabetes. If left untreated, there is a risk of irreversible vision loss or blindness. Generally speaking, the early stage of the lesion does not cause visual impairment, so it is difficult for the patient to detect it, and the best time to see a doctor may be missed, which will lead to the deterioration of the condition. Currently, photographing the fundus can effectively screen for retinopathy. Therefore, regular large-scale screening is necessary. However, there is still a shortage of experienced map-reading doctors, which hinders large-scale screening. In recent years, deep learning has achieved ra...

AI Technical Summary

Problems solved by technology

Second, interference unrelated to diabetic retinopathy is easily amplified by convolution and nonlinear operations, which ultimately affects screening results

Third, the distribution of lesion mask annotation data is extremely unbalanced, and the unbalanced data distribution often greatly inhibits the generalization ability of the model

Therefore, it is meaningful and challenging to study image recognition methods for diabetic retinopathy lesions based on deep learning.

Images