Diagnosis and multiclass classification of diabetic retinopathy using enhanced multi thresholding optimization algorithms and improved Naive Bayes classifier.

Early diagnosis is crucial to prevent a diabetic patient from being affected by blindness. Automatic and accurate detection of diabetic retinopathy is essential. A methodology for the detection and classification of diabetic retinopathy is presented in this paper. Data preprocessing methods are used to highlight subtle information to classify DR anomalies accurately. Image-enhancing techniques are used to boost image quality. Following the preprocessing stage, three main procedures are performed: segmentation, feature extraction, and classification. In contrast to brute force methods, metaheuristic algorithms can explore the solution space more quickly and provide precise, ideal solutions. Due to a lack of detailed image data, it is impossible to determine the precise limits based on image segmentation features. Threshold segmentation is the most effective choice for segmenting fundus images since it has benefits, including simple implementation, low computational complexity, and improved performance. A new variant of grasshopper optimization is proposed using the multi-thresholding version. The segmentation using the proposed model gives high accuracy, even for tiny lesions. A total of 41 features were extracted from the segmented fundus images. Finally, the improved Naïve Bayes classifier classifies the various classes of diabetic retinopathy. The proposed methodology was trained and tested over the DIARETDB0, Messidor-2, Eye pacs-1, and APTOS datasets. The improved naive Bayes classifier enhanced the classification of diabetic retinopathy by an accuracy of 99.98% on the APTOS dataset, which was better than the previously existing techniques. The results proved that the improved naive Bayes classifier adequately diagnoses diabetic retinopathy from the retinal fundus images. [ABSTRACT FROM AUTHOR]