Blindness detection in diabetic retinopathy using Bayesian variant-based connected component algorithm in Keras and TensorFlow

The neuro-degenerative eye disease glaucoma is caused by an increase in pressure inside the retina. As second- leading cause of blindness world, if early diagnosis not obtained, this can total blindness. Regarding fundamental problem, there a huge need to create system that function well without lot equipment, highly qualified medical personnel, and takes less time. proposed modeling consists three stages: pre-training, fine-tuning inference. probabilistic based pixel identification (Bayesian variant) predicts severity Diabetic Retinopathy (DR) which diagnosed presence visual cues, such as abnormal blood vessels, hard exudates, cotton wool spots. article combines machine learning, deep methods for image processing predict images. input picture validated using Bayesian variant connected component architecture, brightest spot algorithm applied detect Region Interest (ROI). Moreover, training sample calculated optic disc cup are segmented with fundus photography ranges 0 4 VGGNet16 architecture SMOTE DR stages images model ensemble ResNet Efficient Net produces excellent accuracy score 93 % predicted Kappa coefficient (p < 0.01) 0.755 retina dataset.