Classification and Segmentation of brain tumors from Magnetic Resonance images for the prediction of the overall survival rates

Abstract

Among all types of cancer, brain tumor is considered the deadliest because of their aggressive nature of spreading, thus affecting both children and adults thereby resulting in a low survival rate. Therefore, timely detection and classification of brain tumors are crucial for planning a patient s treatment. Segmentation of the brain tumor is also required to ascertain the severity of the brain tumor growth for probable treatment planning. Magnetic Resonance Imaging (MRI) is a common imaging technique for diagnosing tumors and other diseases, as it provides superior soft-tissue contrast to delineate the normal and abnormal tissues. The main objective of this research is to classify and segment brain Magnetic Resonance (MR) images using different computational techniques to provide better options to radiologists, thereby increasing the likelihood of survival of patients. To achieve the desired objectives, the information set approaches are investigated for both feature extraction and classification. To this end, the representation of uncertainty and certainty in the distribution of pixel intensities of MR images is cashed in. So the thrust is made on feature extraction, classification, and segmentation using the information set theory. The first attempt is made to devise the possibilistic, probabilistic, and texture features from the brain tumor MR images based on the information set concept. The second attempt is made to create the deep information set features from the feature maps of Convolutional Neural Network (CNN) architecture and the reduced input MR image based on the information set concept, thereby extending the scope of the deep learning neural networks to deal with the uncertainty/certainty in the features. The third attempt is made to develop classifiers such as the Hanman-Shannon transform classifier and the Shannon-Hanman transform classifier based on the t-norm error vector between the training and testing feature vectors. newlineThe fourth attempt is made to develop segmentation using deep learning techn