Diabetic retinopathy detection and classification using hybrid feature set

Local contrast enhancement, adaptive threshold, mathematical morphology methods are used for lesion detection. Multiple machine learning classifiers are utilized to evaluate the performance of proposed approach. The proposed method tested at image and lesion level. AbstractComplicated stages of diabetes are the major cause of Diabetic Retinopathy (DR) and no symptoms appear at the initial stage of DR. At the early stage diagnosis of DR, screening and treatment may reduce vision harm. In this work, an automated technique is applied for detection and classification of DR. A local contrast enhancement method is used on grayscale images to enhance the region of interest. An adaptive threshold method with mathematical morphology is used for the accurate lesions region segmentation. After that, the geometrical and statistical features are fused for better classification. The proposed method is validated on DIARETDB1, E ‐ophtha, Messidor, and local data sets with different metrics such as area under the curve (AUC) and accuracy (ACC).

Source: Microscopy Research and Technique - November 17, 2018 Category: Laboratory Medicine Authors: Javeria Amin, Muhammad Sharif, Amjad Rehman, Mudassar Raza, Muhammad Rafiq Mufti Tags: RESEARCH ARTICLE Source Type: research