Multi-Label Glioma Segmentation Post-Treatment: A 3D U-Net Approach

Abstract

Monitoring post-treatment gliomas is one of the most challenging tasks in medical image analysis, mainly due to the complex tissue changes caused by surgical treatment procedures. The BraTS 2024 initiative addresses this critical need by providing the most extensive collection of expert-annotated MRI data for post-treatment gliomas. Building on this opportunity, we propose a multi-label segmentation approach using a 3D U-Net architecture for post-treatment MRI analysis. Our framework aims to precisely delineate four distinct tumor sub-regions: Enhancing Tissue, Surrounding Non-Enhancing T2/FLAIR Hyperintensity, Non-Enhancing Tumor Core, and Resection Cavity. It leverages the volumetric nature of MRI data through 3D convolutions, allowing superior spatial contextual understanding and feature extraction. Preliminary results demonstrate that our framework can accommodate the inherent complexities of post-treatment tissue characteristics, with significant improvements over baseline models. Thus, it provides a solid foundation for automated post-treatment glioma monitoring.