A Deep Learning based Hybrid Model for Classification of Diabetic Retinopathy

Abstract

Diabetic retinopathy (DR) is a complication of diabetes that effects the retina. Swelling of the retinal blood vessels due to excessive sugar in diabetic patients can cause DR, which damages the retina. Diagnosis of DR is tedious and timeconsuming for clinical experts, hence a computer-aided-diagnosis (CAD) tool is required to detect DR automatically. This paper proposes a novel method for detecting multi-class DR using the Xception model and random forest. First, the fundus images are pre-processed with the contrast-limited adaptive histogram equalization (CLAHE) technique to enhance image contrast by removing the embedded noise. This work proposes a hybrid deep convolutional neural network (DCNN) that concatenates the extracted features from various layers of the pre-trained Xception model to improve the performance. The dimensions of the local features are reduced using linear discriminant analysis (LDA). The resulting features are concatenated and utilized for training the random forest for DR classification. The performance of the proposed method is validated on the publicly available APTOS- 2019 database. The experimental results show that the proposed technique is better compared to the state-of-the-art techniques.