Detecting Retinal Nerve Fiber Layer Using Gray Level Co-occurrence Matrix and Machine Learning Approach

The retinal nerve fiber layer (RNFL) is the white scratch part in the fundus image. It could be an indicator of several diseases characterized by depletion of RNFL. This study intends to construct an RNFL method using Gray Level Cooccurrence Matrix (GLCM) as texture features and implementing random forest as the classifier. The approach has four primary steps: defining the region of interest (ROI), forming the area of features extraction, pre-processing, feature extraction, and classification. Forming the ROI was applied through two main steps: the segmentation with K-means clustering and dividing the fundus image into 12 sub-sectors. The image was then turned into a grayscale as part of the pre-processing. Afterward, the features were extracted using GLCM, followed by features reduction with Correlation-based Feature Selection (CFS). The last step was to use the artificial neural network (ANN) method to classify the data. In this study, the RNFL detection method achieved the accuracy value of 96.7^% against the dataset consisting of 40 images.