Performance Evaluation of Hybrid Machine Learning Algorithms for Medical Image Classification

Abstract

Medical imaging is the process of creating images of parts of human body for diagnosis and treatment purposes. These images are collected from traditional X-ray based methods like Mammography and Computed Tomography (CT). Some advanced sources of images include Magnetic Resonance Imaging (MRI) and Positron Emission Tomography (PET). Since large volumes of digital medical images are deposited in repository, it is a humongous task to access these voluminous images. To access the required image representation from data store, a technique called Content Based Image Retrieval (CBIR) is currently in use. Content-based image retrieval (CBIR) is an assembly that can overcome the problem mentioned above as it is based on the visual analysis of contents that are part of the query image. CBIR retrieves the images which are needed based on its visual contents. CBIR includes Feature extraction and Feature matching. In feature extraction, information like colour, texture and shape known as feature vectors are retrieved through various extraction methods. Similarly, in feature matching the extracted features are compared between normal and abnormal images for classification. The major challenge in CBIR is implementing flexible methodologies to process the different images of different characteristics like colour, shape and pattern. At the same time, applications for retrieving images for proper indexing is done through Picture Archiving and Communication Systems (PACS). In this chapter, retrieving medical images from different data stores and performance of various machine learning classifiers such as Support Vector Machine (SVM) and Deep Learning methodology are focussed to improve the classification accuracy.