Detection of Microaneurysm Model Using Kmean Clustering Method, Hough Transform Optimization and CNN

Introduction: Diabetic retinopathy (DR) is a chronic disease that damages the retina due to damage to the small blood vessels caused by diabetes mellitus. This disease is one of the main causes of visual impairment in people with diabetes. Early detection of clinical signs of DR is essential to allow for effective intervention and treatment. Ophthalmologists are trained to recognize DR by examining small changes in the eye, such as microaneurysms, retinal hemorrhages, macular edema, and changes in the retinal blood vessels. Detection of microaneurysms (MA) plays a vital role in the early diagnosis of RD and has become a major focus of research in recent years. Objectives: The objective of this paper is to propose an automated detection of microaneurysms (MA) in retinal fundus images. In this work, a private database consisting of 50 images from Hospital University Sains Malaysia (HUSM) is used to test the performance of the proposed model. The MA detection model is proposed through segmentation with the Kmean clustering method and Hough transform optimization with Particle Swarm Optimization (PSO) to ensure that the object is MA and is counted. Methods: The methodology in this study is at the pre-processing stage, the image is filtered and the contrast is enhanced. Then in the detection stage the image is segmented using the H-maxima technique and kmean clustering, then the segmented microaneurysms are further identified based on the characteristics of the round shape of the microaneurysm using the Hough transform algorithm optimized with PSO. At the feature extraction stage, the PCA method is used and finally CNN is used to classify the detected MA candidates and calculate their number. Results: The classifier performance was evaluated in terms of accuracy, sensitivity, and specificity as well as the number of microaneurysms before and after detection with Hough Transform optimized with PSO. The CNN classifier achieved good performance with an accuracy of 87.34%, a sensitivity of 93.33%, and a specificity of 88.06%. While PSO had a significant impact on the optimization of the Hough transform algorithm in the number of microaneurysm identifications, the number of Hough transform microaneurysms was better after optimization than before through the round shape feature. Conclusions: This study presents a new model for automatic detection of MA using Kmean with Hough Transform optimized with PSO and CNN for classification. The proposed model successfully classifies MA well including its number quickly for real-time implementation.