On Enhancing Brain Tumor Segmentation Across Diverse Populations with Convolutional Neural Networks

TL;DR

The paper addresses the challenge of generalizing automatic brain tumor segmentation across diverse populations by proposing BraTS-GoAT and evaluating a MedNeXt-based segmentation pipeline. The approach uses four MRI modalities to produce three tumor-region outputs, leveraging sliding-window inference, test-time augmentation, and five-fold cross-validated ensembling, followed by component-based post-processing. On unseen validation data, the method achieves an average DSC of $85.54%$ and HD95 of $27.88$, with larger MedNeXt variants providing additional gains and post-processing significantly boosting performance. The work demonstrates the feasibility of robust, scalable brain tumor segmentation across populations and provides code for replication to support broader adoption and benchmarking.

Abstract

Brain tumor segmentation is a fundamental step in assessing a patient's cancer progression. However, manual segmentation demands significant expert time to identify tumors in 3D multimodal brain MRI scans accurately. This reliance on manual segmentation makes the process prone to intra- and inter-observer variability. This work proposes a brain tumor segmentation method as part of the BraTS-GoAT challenge. The task is to segment tumors in brain MRI scans automatically from various populations, such as adults, pediatrics, and underserved sub-Saharan Africa. We employ a recent CNN architecture for medical image segmentation, namely MedNeXt, as our baseline, and we implement extensive model ensembling and postprocessing for inference. Our experiments show that our method performs well on the unseen validation set with an average DSC of 85.54% and HD95 of 27.88. The code is available on https://github.com/BioMedIA-MBZUAI/BraTS2024_BioMedIAMBZ.

Figures (2)

• Figure 1: The MedNeXt mednext architecture. It combines the benefits of CNNs and transformers by designing transformer-inspired ConvNeXt convnext blocks for image segmentation tasks.
• Figure 2: Qualitative result showing median performance on the validation leaderboard.