Retinal Image Analysis for Ocular Disease Prediction Using Rule Mining Algorithms

• Published in: Interdisciplinary sciences : computational life sciences Vol. 13; no. 3; pp. 451 - 462
• Main Authors: Karthiyayini, R., Shenbagavadivu, N.
• Format: Journal Article
• Language: English
• Published: Singapore Springer Singapore 01.09.2021; Springer Nature B.V
• Subjects: Algorithms; Bayesian analysis; Biomedical and Life Sciences; Centroids; Channels; Clustering; Computational Biology/Bioinformatics; Computational Science and Engineering; Computer Appl. in Life Sciences; Computer applications; Conversion; Disease; Disease detection; Eye; Eye diseases; Feature extraction; Health Sciences; Image analysis; Image processing; Image segmentation; Life Sciences; Mathematical and Computational Physics; Matrix algebra; Medical imaging; Medicine; Original Research Article; Predictions; Retina; Retinal images; Statistics for Life Sciences; Theoretical; Theoretical and Computational Chemistry; FCM; Association rule mining; Image processing; Data mining; Clustering
• ISSN: 1913-2751; 1867-1462; 1867-1462
• DOI: 10.1007/s12539-020-00373-9

Medical image processing is now gaining a significant momentum in clinical situation to undertake diagnosis of different anatomical defects. However, with regard to eye diseases, there is no such well-defined image processing technique in medical image analysis. The scope of this study is to automate computer analysis of ocular disease-related retinal images, which may ease the job of ophthalmologists to rule out the diseased condition. In this present work, eye images are subjected for developing a reliable tool for processing the eye retinal fundus images. The primary objective is to effectively probe retinal image data for providing a holistic approach in automatic fundus disease detection and screening to help clinicians in addition with a developed reliable image processing technique combined with a rule-based clustering method for automatic analysis of fundus images in a reduced time frame. More than 400 eye images available in online are examined. The images were preprocessed by grayscale conversion, retinal segmentation, ROI and crop ROI, image resizing, and extraction in RGB channels. Then these images were segmented by NRR from RGB channels, centroids of rows and columns, and NRR to binary image conversion. Then extraction of features like cup to disc area, optic cup area, and NRR calculations prior to measuring ISNT. A unique algorithm named as EARMAM was introduced for the prediction of diseased image from healthy eye image pool is envisaged in this paper. The functional significance of the EARMAM algorithm was compared with other common classification algorithm of current practice such as SVM, naïve Bayes, random forest, and SMO. The results of confusion matrix have shown that there was 93% prediction accuracy which was higher than the predictive values of other algorithms. The above results are discussed with future improvement and application in clinical field.