Brain Tumor Segmentation

Brain tumor segmentation aims to mark tumor boundaries and subregions in MRI scans, such as whole tumor, tumor core, and enhancing tumor, to support treatment planning and monitoring. Common approaches include 3D U-Net-style convolutional networks, Transformers and state-space models, and fusion of multiple MRI sequences; benchmarks such as BraTS are widely used for comparison. Current work also addresses missing sequences, limited labels, efficient models, and adaptation across hospitals and patient populations. A central challenge is that strong benchmark overlap scores do not guarantee reliable boundaries or performance on new sites, so research increasingly evaluates uncertainty, calibration, and cross-domain robustness.