The computerized analysis of retinal fundus images can be used to conduct the early diagnosis of retinal abnormalities such as diabetic retinopathy (DR). The most major issues with DR detection include noise artefacts brought on by inappropriate lighting, the overlap of lesions and blood vessels because of their similar intensities, and missing data brought on by the processing of a huge quantity of data. This research presented a detection and classification of retinal disease using the KM-WS technique. The main processing phases of the developed DR detection model is Filtering, Segmentation of abnormalities, Feature extraction, Optimal feature selection, and Classification. At first, filter the noises using Block-Matching and 3D (BM3D) approach. The next phase performs the optic disc removal, which is carried out by Kirsch with Mask based Watershed Segmentation (KM-WS) is done for segmenting the blood vessels and its removal.Further, the feature extraction phase is started, which tends to extract four sets of features like Shannon entropy, Kapur entropy, Renyi’s entropy, LBP and GLCM. The feature selection method, which chooses the distinctive characteristics with lower correlation, is carried out since it seems that the feature vector is lengthy. Then, using the FESO algorithm, the important features are chosen from the extracted features. As a result, the classifier PDDLSTM classifies the retinal fundus image using the selected features as input. The paper's major goal is to offer an automated DR detection method that is more accurate, uses less memory, and requires less computing time. By assessing measures like sensitivity, specificity, and accuracy, the performance of the proposed approach is verified using the openly available standard dataset DRIVE. The simulation results show that sensitivity, specificity, and accuracy of 0.95, 0.98, and 0.985 are achieved when comparing the performance of the proposed PDDLSTM algorithm with those of other optimisation techniques.

Kirsch with Mask based Watershed Segmentation (KM-WS), Block-Matching and 3D (BM3D), Fisher Egret Swarm Optimization (FESO), Poisson Distributed Deep Long Short Memory (PDDLSTM) network, retinal fundus image and retinal disease classification.