Deep Learning Based Methods for Brain Tumor Segmentation In MRI: A Review

Abstract

Brain tumor segmentation is an essential task in medical image analysis, since precise identification and depiction of tumors directly assist in the study and treatment planning. Conventional segmentation as well as machine learning based techniques are generally troubled with drawbacks like unsatisfactory generalization, dependency on custom-made features, and susceptibility to imaging artifacts. To overcome these weaknesses, deep learning-based methods have shown to be strong alternatives across various measures in the task of brain tumor segmentation in medical scans. In this review, we examine the latest advancements in deep learning-based approaches, with special emphasis on convolutional neural networks (CNNs), encoder–decoder models, attention mechanisms, and hybrid architectures. Additionally, we explore several benchmark datasets and give a correlative performance of models to give an integrated overview of ultramodern methods. Lastly, we address the problems, such as data imbalance, clarity, and computational cost, as well as possible avenues of future research. The objective of this review is to perform as a starting point and help future research in the area of brain tumor segmentation with deep learning.