Hybrid Explainable Framework for Diabetic Retinopathy Classification from Fundus Images

Diabetic retinopathy is the second most common disease leading to partial or complete loss of vision, but its diagnosis and staging can be difficult and require the use of artificial intelligence technologies. The study presents the realization of a prototype of a hybrid intelligent system based on computer vision technologies, artificial neural networks EfficentNetB0 and ResUNet++, explainable artificial intelligence, fuzzy logic for diagnosing diabetic retinopathy on fundus images. The use of Score-CAM explainable artificial intelligence method increases user confidence by using visual explanation. The proposed solution achieves high accuracy results in the fundus image classification and segmentation task: accuracy - 0.97, specificity - 0.94, DSC - 0.88, IoU - 0.86.