Diabetic Retinopathy Detection with Enhanced Vision Transformers: The Twins-PCPVT Solution
Weinan Dai, Chengjie Mou, Jun Wu, Xuesong Ye
Abstract
Diabetic retinopathy (DR) is a severe diabetes complication that impacts a substantial proportion of individuals living with diabetes. It is estimated that around 40-45% of Americans with diabetes experience various stages of this condition. Preventing blindness relies heavily on timely detection and treatment. We use an enhanced vision transformer-based model in the detection of DR, named Twins-PCPVT. The model incorporates a twin architecture that captures both global and local features of fundus images, attaining impressive accuracy and AUC values of 87.43% and 0.952, respectively, on the Kaggle Diabetic Retinopathy Detection dataset. Our proposed method demonstrates great potential in aiding early diagnosis and treatment of DR. Our proposed deep learning-based approach offers a notable advantage over the current laborious and time-consuming manual detection process. It is not only faster but also more efficient. By implementing our method, early detection and treatment of DR can be significantly improved, playing a crucial role in preventing vision loss.