Detection of Diabetic Retinopathy in Retinal Fundus Images Using CNN Classification Models

• DOI: 10.3390/electronics11172740
• Publication Date: 2022-09-01T03:53:21Z
• Authors: Al-Omaisi Asia; Cheng-Zhang Zhu; Sara A. Althubiti; Dalal Al-Alimi; Ya-Long Xiao; Ping-Bo Ouyang; Mohammed A. A. Al...
• Keywords: Residual neural network; Fundus (uterus)

Diabetes is a widespread disease in the world and can lead to diabetic retinopathy, macular edema, other obvious microvascular complications retina of human eye. This study attempts detect retinopathy (DR), which has been main reason behind blindness people last decade. Timely or early treatment necessary prevent some DR control blood glucose. very difficult time-consuming manual diagnosis because its diversity complexity. work utilizes deep learning application, convolutional neural network (CNN), fundus photography distinguish stages DR. The images dataset this obtained from Xiangya No. 2 Hospital Ophthalmology (XHO), Changsha, China, large, little labels are unbalanced. Thus, first solves problem existing by proposing method that uses preprocessing, regularization, augmentation steps increase prepare image XHO for training improve performance. Then, it takes advantages power CNN with different residual (ResNet) structures, namely, ResNet-101, ResNet-50, VggNet-16, on datasets. ResNet-101 achieved maximum level accuracy, 0.9888, loss 0.3499 testing 0.9882. then assessed 1787 photos HRF, STARE, DIARETDB0, databases, achieving an average accuracy 0.97, greater than prior efforts. Results prove model (ResNet-101) better ResNet-50 VggNet-16 classification.