Binary Mode Multinomial Deep Learning Model for more efficient Automated Diabetic Retinopathy Detection

The ability to rapidly and accurately classify diabetic retinopathy from color fundus photographs is vital to maximize the ability to assess diabetic eye disease early. Our paper compares the performance of binary classification (Refer/No Refer) to multinomial (Diabetic Retinopathy Severity) classification using deep learning models. The binary mode multinomial experiment achieved very high performance for Refer/No Refer DR on clinical datasets with accuracy up to 97.69%. We show how annotating images using image processing improves Multinomial classification in binary mode on a set of fundus images and yields equal if not better performance than a simple binary classification on the same dataset.