Brain Tumor Segmentation in mpMRI Scans (BraTS-2021) Using Models Based on U-Net Architecture

Accurate segmentation of brain tumors from MR images has important clinical relevance. To overcome the limitations of manual segmentation, a semi-automatic or automatic approach is desirable. The current study mainly focused on task-1 defined in the BraTS'21 challenge, i.e., segmenting glioblastoma into sub-regions (the "enhancing tumor" (ET), the "tumor core" (TC), and the "whole tumor" (WT)). In the current study, deep learning models based upon UNet architecture were developed for producing tumor segmentation labels from 3D multi-parametric MRI (mpMRI) scans. During the segmentation process, the output mask resulting from the first developed Model (WT) was used to develop segmentation models for the remaining subregions. Optimizations were carried out to develop robust models, reduce computation time, and achieve high accuracy. Developed models showed high accuracy for the segmentation of tumor sub-regions on training as well as validation data.