Multiscale and Texture-Aware Deep Learning for Brain Tumor Segmentation in MRI: A Systematic Review

Abstract

The development of aberrant brain cells, some of which may be malignant, is known as a brain tumor. For patients to survive, brain tumors must be identified and diagnosed early. The tumor segmentation stage in the analysis of brain tumor images is essential to the further processing of MRI images. The tumors designated for segmentation are anatomical structures that are often characterized by flexibility and complex morphology, exhibiting significant variability in their location and structure, and differing from patient to patient. Accurate brain tumor segmentation faces challenges such as location uncertainty, morphological uncertainty, low contrast imaging, annotation bias and data imbalance. The aim of this systematic literature review is to determine which deep learning segmentation technique employing multiscale texture analysis is suitable for segmenting the uneven, diffuse tumor boundary from the adjacent brain matter. We conducted a systematic review of 970 brain tumor segmentation studies which employed multiscale features, texture features and deep learning techniques. Also, segmentation techniques using T1-weighted, T2-weighted, gadolinium-enhanced T1-weighted, fluid-attenuated inversion recovery, diffusion-weighted and perfusion-weighted magnetic resonance imaging sequences were reviewed. We synthesized each method as per the deep learning method used, strengths and weaknesses of the proposed technique, dataset used, types of brain tumors analyzed, feature types employed, and the performance measures achieved. It was observed that, T1-weighted, T2-weighted, gadolinium-enhanced T1-weighted, and fluid-attenuated inversion recovery MRI are the most used in the different segmentation algorithms. We found that multiscale and texture features have emerged as critical components in improving the accuracy and robustness of brain tumor segmentation from MRI images. Incorporating these features addresses the inherent challenges posed by the complex nature of brain tumors, including variations in size, shape, location and blurred edges.