An effective technique for diabetic retinopathy using hybrid machine learning technique

Abstract

Diabetic retinopathy (DR) stays as an eye issue that has continuously developed in individuals who experienced diabetes. The complexities in diabetes cause harm to the vein at the back of the retina. In outrageous cases, DR could swift apparition disaster or visual impairment. This genuine impact had the option to charge through convenient treatment and early recognition. As of late, this issue has been spreading quickly, particularly in the working region, which in the end constrained the interest of an analysis of this disease from the most prompt stage. Therefore, that are castoff to protect the progressions of this disorder, revealing of the retinal blood vessels (RBVs) play a foremost role. The growth of an abnormal vessel leads to the development steps of DR, where it can be well known by extracting the RBV. The recognition of the BV for DR by developing an automatic approach is a major aim of our research study. In the proposed method, there are two major steps: one is segmentation and the second one is classification of affected retinal BV. The proposed method uses the Kinetic Gas Molecule Optimization based on centroid initialization used for the Fuzzy C-means Clustering. In the classification step, those segmented images are given as input to hybrid techniques such as a convolution neural network with bidirectional-long short-term memory (CNN with Bi-LSTM). The learning degree of Bi-LSTM is revised by using the self-attention mechanism for refining the classification accuracy. The trial consequences disclosed that the mixture algorithm achieved higher accuracy, specificity, and sensitivity than existing techniques.