TwinTrack: Post-hoc Multi-Rater Calibration for Medical Image Segmentation

TwinTrack is a post-hoc calibration method that aligns ensemble segmentation probabilities with the average label fraction across multiple expert raters.

Applied to pancreatic ductal adenocarcinoma contrast-enhanced CT segmentation, it improved calibration on a multi-rater benchmark.

The method makes segmentation probabilities more interpretable when experts disagree, better reflecting real diagnostic uncertainty.