Résumé: Diabetic retinopathy (DR) is a kind of eye disorder that injures retinal blood vessels and finally leads to complete visual impairment. The severity of eye conditions can be avoided by taking preventive measures like proper screening, early detection, and timely diagnosis. Nowadays, automated diagnostic systems are developed by researchers to monitor the progression of eye disease. The automated systems use machine learning techniques to discriminate retinal fundus images based on the severity grading. The methodologies currently being used in this domain, however, have several limitations on DR grading tasks, including redundant training time, the use of data annotation, and reduced sensitivity performance. Therefore, we proposed a novel automated diagnostic system using the Komodo Mlipir Algorithm-optimized Attention Bidirectional Long Short-Term Memory (KMA-optimized ABiLSTM) approach to detect the symptoms of DR in the early stage from the retinal fundus images. The retinal fundus images used to evaluate the proposed method were obtained from the MESSIDOR and IDRiD databases. The proposed KMA-optimized ABiLSTM technique detects and categorizes DR abnormalities in datasets as non-DR, mild non-proliferative DR (NPDR), moderate NPDR, severe NPDR, and proliferative DR (PDR). The effectiveness of the proposed KMA-optimized ABiLSTM approach is examined by comparing its performance with other state-of-the-art approaches in terms of diverse performance measures. The experimental analysis inherits that the proposed KMA-optimized ABiLSTM approach achieves a high accuracy rate of about 98.5% with less computation time (50 s) than other state-of-the-art approaches.

Langue: Anglais