Recent Segmentation Trends of MRI-Based Brain Neoplasm: A Review

Abstract

A significant increase in medical-related cases of brain neoplasm has been noticed in the past few years affecting not only adults but children as well. Brain neoplasm segmentation isolated the different abnormal brain tissues from normal brain tissues which are complicated tasks in medical image examination. But simultaneously it plays a crucial role in time diagnosis which not only improved the treatment possibilities but also increased the rate of survival of the patients because a brain neoplasm is a treatable kind of cancer if diagnose well on time. Magnetic resonance imaging (MRI) based segmentation are a more focused, attractive, and attentive study area in recent few years because MRI is noninvasive imaging, safe, and cost-effective. Detection of brain neoplasm using the manual procedure of segmentation is an extremely difficult, time-consuming, expensive, and individual task because large data of MRI images are produced in clinical practice which may delay the diagnosis. This increases the practical significance of the automatic segmentation techniques but due to brain neoplasm being extremely unpredictable concerning position, appearance, type, and size improving the methods for state-of-the-art segmentation process remain a complex task. To segment the brain neoplasm accurately, automatic segmentation is a solution with better performance. The goal of this paper provides an organized literature survey for recent MRI-based automatic brain neoplasm segmentation techniques with the modern participation of several researchers which helps new researchers in exploring future directions. There are several current surveys focused on traditional methods, but this paper focuses on recent trends including machine learning techniques accompanied by transfer learning, deep learning, neural network, and hybridization. Moreover, this survey also presents the findings and limitations of each article which show the effectiveness of the proposed work. Finally, this survey, found that after over two decades of research, the novel methods for segmenting brain neoplasms using computer-aided techniques are becoming more and more refined and becoming closer to being used often in clinical settings but in terms of computing complexity and memory consumption, these approaches lag.