Automated Brain Tumor Segmentation in MRI Images Using Deep Learning: Overview, Challenges and Future

Abstract

Brain tumor segmentation of MRI images is a crucial task in the medical image processing. It is very important that a brain tumor can be diagnosed in initial stages which eventually improve treatment as well as survival chances of patient. Manual segmentation is highly dependent on doctor, it may vary from one expert to another as well as it is very time-consuming. On the other side, automated segmentation helps a doctor in quick decision making, results can be reproduced and records can be maintained electronically which improves diagnosis and treatment planning. There are numerous automated approaches for brain tumor detection which are popular from last few decades namely Neural Networks (NN) and Support Vector Machine (SVM). But, recently Deep Learning has attained a central tract as far as automation of Brain tumor segmentation is concerned because deep architecture is able to represent complex structures, self-learning and efficiently process large amounts of MRI-based image data. Initially the chapter starts with brain tumor introduction and its various types. In the next section, various preprocessing techniques are discussed. Preprocessing is a crucial step for the correctness of an automated system. After preprocessing of image various feature extraction and feature reduction techniques are discussed. In the next section, conventional methods of image segmentation are covered and later on different deep learning algorithms are discussed which are relevant in this domain. Then, in the next section, various challenges are discussed which are being faced in medical image segmentation due to deep learning. In the last section, a comparative study is done between various existing algorithms in terms of accuracy, specificity, and sensitivity on about 200 Brain Images. The motivation of this chapter is to give an overview of deep learning-based segmentation algorithms in terms of existing work, various challenges, along with its future scope. This chapter deals with providing the crux of different algorithms involved in the process of Brain Tumor Classification and comparative analysis has also been done to inspect which algorithm is best.