Diabetic Retinopathy Detection Using SOBA Machine Learning Framework

Abstract

Computer vision-based image classification for disease proliferation or possibility of prognosis is an important approach and becomes one of the major needed tasks in the medical industry. Having such significant thrust area attracted the research studies here and triggered to address the problem of identifying the Diabetic Retinopathy (DR). As more research work surfaced out recently with encouraging results a novel framework SOBA is proposed and verified with unique combination of image processing algorithms of first of its kind and the four components are designed as follows. Firstly, the S-Aspect denotes the models in shallow learning based on architecture with few layers in a neural network or few levels in a decision trees/Rules and probabilistic networks. Secondly the O-aspect as Orchestration of Deep learning architectures. Thirdly the B-Aspect denotes balancing the class distribution and finally the A-Aspect as the attribute reduction. This framework can be considered as basis for the selection of pre-processing methods and transforming the data feasible for classification for in categorizing the input image based on the level of diabetic retinopathy as the central issue of this research study. SOBA framework enables to achieve results firstly the better accuracy of 70.15% and Weighted Average ROC values 0.862 using first component, and while applying the second component the accuracy improved to 81.99%, Weighted Average of Receiver Operating Characteristics (ROC) 0.907. The third component yields confirming similar performance of accuracy of 81.75% and weighted average ROC 0.905.the fourth component achieved better accuracy 73.60%. in the model. Finally, the fifth component yield very high accuracy rate achieved 82.29%. More over these percentage and weighted average ROC 0. 905. results summaries with a figure 81.99% for conventional machine learning and a figure 82.29% for deep learning. newline newline newline