Diabetic retinopathy feature extraction images based on confusion neural network

Abstract

The diagnosis of diabetic retinopathy depends on the evaluation of retinal fundus pictures. The current methods have been successful in extracting features from fundus images, but due to the complex blood vessel distribution in these images and the presence of a great deal of noise, simple methods based on threshold segmentation and clustering are vulnerable to feature loss during the extraction process. For example, the small blood vessels in the fundus are lost, and the branches of blood vessels are blurred. In addition, the noise in medical images is mainly distributed in the high-frequency area of the image. The proposed method to segment the retinal fundus vessels in the DRIVE and STARE datasets, the average accuracy of this method is 95.45% and 94.81%, respectively, and the sensitivity and specificity are 73.35%, 75.39% and 97.34%, 95.75%. In addition, compared with related methods, the proposed method has higher segmentation accuracy, and after segmentation, the fundus blood vessels have higher integrity, clear structure, and less loss of small blood vessels.