Early detection of DR with an Effective Optimal Stochastic Deep Network in fundus images using the Monte Carlo Method

Diabetes mellitus often causes diabetic retinopathy (DR), which damages the retina and makes it hard to see. If it is forced to treat, the possibility of losing sight is much less likely. It can cause hallucinations if not diagnosed early. Manually examining DR retina fundus pictures is a time-consuming, labor-intensive process that is also prone to error for ophthalmologists, making it preferable to CAD technology. Deep learning (DL) has notably become one of the most prominent ways to improve effectiveness, especially when it comes to putting clinical data into categories and evaluating them. On the other hand, reliability limits the incorporation of DL into real-world medical processes since traditional DL frameworks cannot objectively evaluate model uncertainty. The Enriched Squawk Optimization Algorithm (ESOA) is utilized to fine-tune the hyperparameters of SDNMC and is referred to as OSDNMC in this study. This method, which can estimate uncertainties to assess tumor image classification reliability, is proposed as a solution to this problem. This model analyses the concept using fundus image datasets collected with the customized device from high-risk populations to demonstrate that actionable uncertainty information may be generated. The tests also demonstrate better accuracy using uncertainty-informed recommendations. By employing color fundus images and the OSDNMC technique, our work was able to classify DR automatically with 96.5% accuracy on our dataset, exceeding findings from more traditional methods like Ar-HGSO, CNN, and DCNN.