Automated Multiclass Diagnosis of Diabetic Retinopathy using Hierarchical Learning

Abstract

Diabetic Retinopathy (DR) is the leading cause of blindness in the modern world. Diagnosis of DR requires an experienced ophthalmologist and it is a tedious and time-consuming process. In this paper, we propose a Convolutional Neural Network (CNN) based automated diagnosis system that can classify various stages of DR accurately. A hierarchical approach is adopted for classification in which we break down our classification task into two stages. In the first stage we perform binary classification obtaining the true positive and negative samples and in the second stage, five class classification is performed on those images classified as true positive, false positive and false negative in the first stage of classification. Our approach of classifying hierarchically takes care of the class imbalance in the data by removing the most dominant class 0 (No-DR) from the dataset at the binary classification stage. The proposed method uses the Inception-v3 CNN for feature extraction in which we use the features from second last layer of both main and auxiliary classifiers. The extracted features are concatenated into a single feature vector to train a Support Vector Machine (SVM). We use SVM with Radial Basis Function (RBF) kernel for both binary and multiclass classifications. Experiments are conducted on Kaggle dataset and our approach attains an accuracy of 87.7% on validation data for binary classification and 81.8% for multiclass classification. Our results are better than the recently proposed approach using CNN indicating that the hierarchical classification performs better for multiclass classification.