GLAUCOMATOUS IMAGE CLASSIFICATION CAD SYSTEM USING ADAPTIVE WAVELETS, PROBABILISTIC PCA AND RANDOM FOREST TECHNIQUES MACHINE LEARNING MODEL

Abstract

World Health Organization (WHO) in a recent study has identified eye related defects to be one of the primary health challenges faced by the current society. Common retinal abnormalities include Glaucoma, Diabetic retinopathy and Macular degeneration[16]. Retinal eye defects have significantly increased in the last decade across developing and developed countries. Retina is a critical part of the eye enabling conversion of light to neural signals to enable the brain for visual recognition. Glaucoma disorder causes damage to human’s optic nerve due to the increased pressure in the eye. The optic nerveis very important in the eye as it transmits the images/signals to the brain. Glaucoma leads to permanent vision loss if the optic nerve damages increases continuously. Hence, an early diagnosis is required to prevent vision loss. In this paper, an efficient method for glaucomatous image classification is presented using Dual Tree M-band Wavelet Transform (DTMWT), Probabilistic Principal Component Analysis (PPCA) and Random Forest (RF) classifier. At first, DTMWT is applied to represent the fundus image in multi-resolution that contains lower and higher frequency components. The lower frequency components are reduced by PPCA and then classification is made by RF classifier. The performance of the system is measured in terms of accuracy, sensitivity and specificity. Results show that a maximum classification accuracy of 91%, sensitivity of 88%, and specificity of 94% are obtainedby PPCA based DTMWT features with RF classifier.

Keywords: Glaucomatous images, DTMWT, PPCA, RF classifier, Cancer System, CAD System, Medical Imaging, Image Processing, Performance Evaluation, Wavelet Transform, Machine Learning, Image Data, Medical Databases.