An efficient detection of micro aneurysms from fundus images with CDLNN algorithm

• Published in: Materials today : proceedings 2023, Vol.81, p.553-562
• Main Authors: Sandhya, S.G., Suhasini, A.
• Format: Article
• Language: English
• Subjects: Context unit based Deep Learning Neural Network (CDLNN); Gaussian Contrast limited adaptive histogram equalization Model (GCM); Pearson's Correlation-based Local Topography (PCLT); Weighted Chicken Swarm Optimization (WCSO)
• ISSN: 2214-7853; 2214-7853
• DOI: 10.1016/j.matpr.2021.04.010

Diabetic Retinopathy (DR) is an eye disease that damages or affects the retina and is caused on account of diabetes. It might bring about blindness in the future. Detection as well as classification of DR at an early stage could significantly lessen the severity of this vision loss. Existing researches are time-consuming and they show less accuracy. Hence, for enhancing the DR system’s accuracy, this paper proposed an enhanced DR detection and classification system utilizing CDLNN. The proposed methodology comprises the succeeding phases. Primarily, the image is inputted to the pre-processing phase utilizing the GCM algorithm. Subsequently, segmentation of blood vessels and some affected injuries, namely Cotton Wool Spots (CWS), microaneurysms (MAs), exudates, and hemorrhages, are done utilizing the PCLT algorithm and mask unit, respectively. Also, these injuries are checked in the segmented blood vessel (BV) utilizing the same mask unit. After that, the SIFT, LTrP, color histogram, edge, shape, kurtosis, Shannon’s entropy, and homogeneity features are well-extracted as of the segmented injuries. Subsequently, important features are selected with the utilization of the WCSO algorithm. Next, the selected features are inputted to the CDLNN classifier, which classifies the retina image as, normal, PDR, and NPDR. During an experimental evaluation, the proposed system attains excellent performance on considering the existing ones.