Diabetic Retinopathy Classification via Downscaling and Deep Learning

A new arXiv preprint proposes using downscaling algorithms before feeding retinal images into a deep learning network for Diabetic Retinopathy (DR) classification. DR classification involves grading fundus images into five severity stages. The authors combine two datasets—Kaggle and the Indian Diabetic Retinopathy Image Dataset—to improve training and testing. They employ a novel Multi Channel Inception V3 architecture with a custom preprocessing phase. Results show improved accuracy, specificity, and sensitivity over previous state-of-the-art methods. The paper addresses the challenge of large and varying image sizes in DR classification.