A Novel Transformer Model with Multiple Instances Learning for Diabetic Retinopathy Classification

A Novel Transformer Model with Multiple Instances Learning for Diabetic
Retinopathy Classification

Authors- S Arun Prasath, Assistant Professor Dr.P.Kavitha

Abstract-Diabetic retinopathy (DR) is a major cause of visual impairment worldwide, and hence, early and correct detection is needed to avoid severe consequences. This project proposes a new transformer-based model combined with Multiple Instance Learning (MIL) for diabetic retinopathy classification from retinal fundus images. The transformer model encodes long-range dependencies and contextual information, whereas the MIL framework handles image patches to concentrate on diagnostically important areas. This is a hybrid methodology that provides stable feature extraction and classification in diverse image resolutions and noise levels. The model is trained on an extensive dataset and proves to have higher sensitivity and specificity than standard deep learning practices. The system seeks to assist ophthalmologists in providing accurate and timely diagnoses, providing an efficient and scalable solution for DR screening programs on a large scale.

DOI: /10.61463/ijset.vol.13.issue3.116