A Data Driven Approach for Automated Brain Tumor Segmentation & Classification

Machine Learning and Deep Learning are finding their applications in several domains, medical and biomedical applications being one of the most critical among them. A lot of research has gone into development of data driven models for cancer detection to serve as an id for medical practitioners. Brain cancer happens to be one of the most challenging forms of cancer to detect at an early stage as part of the tumor needs to be extracted from the brain for the biopsy analysis, which further decides the direction of treatment. As malignant and benign tumours have different treatment protocols, it is of utmost importance to detect and segment brain tumours accurately at the outset to ensure successful treatment and minimize chances of mortality. This paper presents a data driven approach for segmentation and subsequent classification of brain cancer datasets based on machine learning and deep learning approaches. The machine learning model employs statistical feature extraction followed by classification using a neural network model. Image filtration is employed prior feature extraction to circumvent potential effects of noise and blurring. The deep learning models employed are the multi-instance learning (MIL) and the residual network (ResNet) models. The classification accuracy of the models is 97.5, 98% and 99% for the Neural Network, MIL and ResNet models respectively.