Dendritic Learning Empowered Vision Transformer for Diabetic Retinopathy

Abstract

Deep learning has revolutionized complex tasks such as classification, approximation, and prediction, drawing inspiration from mathematical models of the human brain. Among recent breakthroughs, Google's Transformer architecture has established itself as a leading framework in natural language processing. Its adaptation to computer vision, known as the Vision Transformer (ViT), has set new benchmarks for image-based tasks. In this study, we introduce a novel neural network model that integrates the input layer of the ViT with the dendritic neuron model (DNM). This hybrid architecture combines the advanced feature extraction capabilities of ViT with the adaptability and robustness of DNM to enhance performance. The proposed model is applied to the diagnosis of diabetic retinopathy, effectively identifying critical features associated with the condition. The results underscore its potential to improve the accuracy and reliability of medical image analysis, paving the way for advancements in healthcare diagnostics.