tumor detection and analysis in brain mir images using machine learning techniques

Abstract

Magnetic resonance imaging (MRI) of brain is a non-invasive medical imaging method mainly used to measure and visualize anatomical structure, abnormalities of human s brain.MRI has been effectively used in examining other internal body structures like strokes, tumors, sclerosis, aneurysms, eye, cysts, breast cancer, inner ear, and many more. In human body, normal cells grow in a controlled manner with new cells replacing the old or damaged cells. On other hand, tumor cells grow in an uncontrolled manner. Further, brain tumor in simple words can be defined as the abnormal growth of the cells inside the human brain. newlineIn traditional method, the radiologists diagnose lesions present in the brain by looking into the neuro images manually. But, in practical cases, this process is very time-consuming and human error subjected. Hence, it is recommended to employ automated methods to perform brain lesion detection which would definitely assist doctors in improved diagnosis. This way, the goal of this thesis is to develop automated machine learning and deep learning methods for delineating brain lesions in MRI scans. Hence, in this thesis, we propose different computer aided diagnosis (CAD) systems based on machine learning and deep learning algorithms to effectively detect brain lesions in MRI images. The proposed cascaded machine learning based brain lesion (CML-BR) detection method is based on employing machine learning techniques for abnormality detection and severity assessment. The proposed CML-BR method employs different descriptors to extract features which perform as detection factor of brain abnormalities, the obtained features are then concatenated to obtain a discriminative feature vector. Further, another proposed Convolution Neural Network based brain lesion (CNN-BR) detection method is based on employing deep learning approach for abnormality detection. The proposed CNN-BR method uses CNN architecture and transposed convolution as the base function. Hence, a robust architecture is used to segment brain lesion