Recognition and classification of medical images using machine learning approaches

Abstract

newlineComputer Aided Diagnosis is becoming popular in medical sciences as it provides newlineaccuracy and timeliness, the two major aims of medical field. In the thesis presented newlinehere, an algorithm is developed which aims to design an auto-CAD system for the newlinediagnosis of retina abnormalities. Diabetic Retinopathy becomes severe if not newlinediagnosed and treated at the first stage. Age-related Macular Degeneration is another newlinevision threatening disease that occurs in the elderly population and needs serious newlinemedical attention. In this research work, these two diseases are considered and the newlinesigns of these two diseases are analyzed. A combined database is formed by collecting newlinethe images from several standard datasets. The algorithm presented in this thesis is newlinedeveloped with the combination of two steps, namely, image processing and machine newlinelearning. Several image processing algorithms for segmentation and morphological newlineoperations are used for the detection of the abnormalities caused by the above newlinementioned diseases. A set of significant features are selected and evaluated on the newlineabnormalities extracted in the image processing stage. The classification of the newlineabnormalities with a training and a test set is performed using different machine newlinelearning algorithms. The random forest classifier is best suited to the dataset used in newlinethis research for its performance accuracy and robustness with respect to noise. With newlinethe aim of forming a Case Based Reasoning model, we have developed a method of newlinemachine learning based classification of different abnormalities. In future studies, an newlineauto-CAD system with Case Based Reasoning paradigm is aimed to be developed newlinedepending on Content Based Image Retrieval model.