Toward Enhanced Brain Tumor Segmentation in MRI: An Ensemble Deep Learning Approach

Convolutional neural networks (CNNs) have played a pivotal role in enhancing brain tumor segmentation techniques. In this research, we employed an integrated approach using three distinct CNN architectures: U-Net, SegNet, and DeepLabv3+, to segment brain tumors from MRI scans. The integration of these models' outputs was accomplished through a smart majority voting technique (SMVT), which considers the accuracy levels of the individual networks. Our evaluation on the high-grade glioma (HGG) subset from the 2018 BraTS challenge revealed impressive dice similarity coefficients: 0.92 for the enhancing tumor (ET) region, 0.94 for the tumor core (TC), and 0.96 for the whole tumor (WT) area. These findings not only exhibit superior performance over several existing approaches but also underscore the effectiveness of ensemble deep learning models in medical image analysis.