Diabetic Retinopathy Blood Vessel Detection Using Deep-CNN-Based Feature Extraction and Classification

Abstract

Diabetic Retinopathy (DR) is the main cause of blindness and harms the retina due to the accumulation of glucose in the blood. Therefore, early DR detection, diagnosis, segmentation, and classification prevent patients with diabetes from losing their vision. However, the main challenge is the test takes time, lots of procedure, and money for proliferative stage of DR and identifying the proper glucose level that is present in the blood vessel. To address this problem, a unique and hybrid method, namely Deep Convolutional Neural Network (CNN) with EFficientNetB0 (Deep-CNNEF-Net), is used for earlier detection and classification of DR. Furthermore, the proposed scheme follows the pre-processing, input feature extraction, and classification phases. The pre-processing stage improves the presence of abnormalities and segmentation of DR by using the CNN with mean orientation technique. Moreover, the input feature extraction step uses CNN to obtain features that are important for training purposes. Furthermore, a novel modification to the VGG16 pooling layer uses the Global Average Pooling Layer (GAP) without a fattening layer for the boundary box of infected retinal blood vessels. The experimental results are obtained using two databases, namely the Digital Retinal Images for Vessel Extraction (DRIVE) and the STructured Analysis of the Retina (STARE) datasets, for the classification and detection of DR. The simulation results have been examined by evaluation metrics, such as accuracy, precision, recall, F1-score, specificity, and Mean Process Time per Frame (MPTF) for proper validation. The evaluation analysis outperforms the existing State-Of-The-Art (SOTA) models on the same eye retinal DR datasets.