Ensemble Models for Multi-class Classification of Diabetic Retinopathy

Artificial intelligence-based methods help in the automatic detection and diagnosis of various diseases in the medical field. Diabetes affects 5 in 10 people in the world. It can cause diabetic retinopathy(DR) that damages the retina, leading to irreversible vision loss. DR is generally diagnosed by an Ophthalmologist using a patient's fundus image. The Ophthalmologist grades the disease based on severity. Early intervention can delay the progression of the disease. Automatic disease detection can help large-scale patient screening, early detection of diabetic retinopathy, and reduce human error. The work aims to study various ensemble learning models and their ensemble voting methods for the DR classification task. The study also focuses on the effects of data augmentation along with preprocessing. The performance of each of the models has been studied using the Aptos-blindness detection 2019 dataset.