Paper Title
An Interpretable Hybrid CNN–Fuzzy Approach for Multi-Class Diabetic Retinopathy Classification

Abstract
Diabetic retinopathy (DR) is considered one of the main reasons for global blindness, thus requiring effective automated techniques for diagnosis. The Convolutional Neural Network (CNN) models based on deep learning have proved successful in classification tasks for retinal images. However, the lack of transparency limits the usage of CNN models for medical applications. This paper suggests the CNN-Fuzzy model, which is a combination of a deep neural network and fuzzy systems. First, features are extracted using a pretrained ResNet50 model and then they can be quantized through PCA, which will be further classified by a fuzzy inference system. CNN predictions and fuzzy predictions are fused using a weighted adaptive fusion scheme. It has been shown that the proposed hybrid approach provides 0.76 accuracy and Macro F1 score of approximately 0.48, which is comparable to the baseline CNN approach. Keywords - Diabetic Retinopathy, CNN, Fuzzy Logic, Hybrid Model, PCA, Medical Image Classification