Confidence is Not Reliability: Rethinking MC Dropout in Brain Tumour Segmentation
Researchers have investigated the reliability of Monte Carlo Dropout (MC Dropout) for segmenting brain tumors in MRI scans, finding that while it can align uncertainty with errors, it may not always guarantee clinical safety. In a study using 126 BraTS21 patients, MC Dropout demonstrated strong uncertainty-error alignment, correctly ranking erroneous voxels higher and identifying subgroups with significantly lower segmentation performance. However, the study also revealed that global alignment metrics can mask critical region-specific calibration failures, as seen with one model exhibiting severe miscalibration on a clinically vital sub-region despite a high overall AUROC score. The findings emphasize the need for sub-region-specific calibration assessments alongside standard metrics when selecting models for clinical deployment. AI
IMPACT Highlights the need for more robust uncertainty quantification in medical AI to ensure patient safety and reliable clinical deployment.