Regression Model-based Feature Filtering for Improving Hemorrhage Detection Accuracy in Diabetic Retinopathy Treatment

Abstract

Diabetic retinopathy (DR) is an optical syndrome infecting the eyes’ vision by impairing the retinal blood vessels. Early misdetection of impairment results in hemorrhage, a state in which retinal bleeding occurs. Therefore, initial detection of such bleeding in the retina is identified using intelligent computing and clinical analysis. This analysis helps to improve the precision of detection and requires complex-less time and processing instances. In this article, the regression model for retina feature filtering (RM-FF) is introduced to improve the accuracy of detecting hemorrhages. In this filtering, the complex image is simplified into smaller blocks for classification and conditional verification. Based on conditional verification, the training set is updated recursively to improve the specificity and sensitivity detection process. Using a differential dataset, the proposed detection method assessed using the metrics true positive rate, accuracy, sensitivity, and specificity.