Optic Disc and Blood Vessel based Diabetic Disease Detection Using Enhanced Long-Short-Term Memory

Diabetic retinopathy (DR) is an eye disorder caused by diabetes. In this disease, progressive damage to the retina occurs if high blood glucose levels are not controlled. Early diagnosis and treatment can reduce the risk of vision loss. Retinal fundus image plays a significant role in early DR classification and identification. Creating an accurate diagnostic model using retinal fundus imaging depends on image quality and size. This work proposes a deep learning based optic disc and blood vessel disease segmentation and classification technique. Initially, retinal images are collected from publicly available datasets and the data is pre-processed. Then, the optic disc and blood vessel images are segmented using U-net. After that, the contrast and similarity features of the segmented image are extracted using a gray-level co-occurrence matrix (GLCM). The segmented image is then passed as input to enhanced long-short-term memory (ELSTM). Here, retinal image classification is achieved by LSTM technique. The weights and biases of the LSTM are optimized using a tree seed optimization algorithm. Retina dataset from RFMiD is used to validate the proposed work. The proposed method is implemented in Python. From the experimental results, our proposed approach achieved a high accuracy of 97.89% compared to the state-of-the-art methods in terms of various metrics and performance.