Diabetic Retinopathy Fundus Image Classification Using MSCSO Based Optimization with Fuzzy Support Vector Machine

Complications of diabetes can lead to a disease known as diabetic retinopathy (DR), which causes irreversible damage to the vessels in the retina. If not caught in its early stages, DR is a major source of preventable blindness. Steady scanning with high-efficiency computer-based schemes for early diagnosis is crucial due to the fact that current DR treatments can only halt or delay the worsening of vision. Fully automatic diagnosis systems were presented in this work that outperform manual methods in preventing incorrect diagnoses and saving time and money. In addition to pinpointing the precise location of affected lesions on the retain surface, the proposed system categorises DR imageries into five distinct stages: no DR, mild DR, reasonable DR, Spartan DR, and DR. Accordingly, this work provides an effective image processing strategy for diagnosing DR disorders from retinal fundus pictures, one that can reach the necessary presentation metrics (automatic diabetic retinopathy screening includes preprocessing, detection and removal of the optic disc, segmentation and removal of blood vessels, removal of the fovea, extraction of features (including microaneurysms, retinal haemorrhages, and exudates), classification, and finally, screening. In this research, we offer a new method for feature selection based on a variant of the sand-cat swarm optimization (MSCSO) procedure for eliminating irrelevant information. Nomadic spanning tree search and optimisation (MSCSO) is a method that takes this technique. To get an upper hand in an encounter, a sand cat may move around until it locates an appropriate perch. Using the MSCSO algorithm with a wandering approach improves the sand cat's mobility and the algorithm's ability to conduct global exploration. We then implement the lens based learning technique to increase the convergence speed of the algorithm. In this instance, a fuzzy support vector machine (FSVM) is used for classification. The system is evaluated versus the state of the art using APTOS Kaggle 2019 public datasets. Experiments showed a high level of sensitivity, specificity, and accuracy.