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

Tristan Kirscher, Alexandra Ertl, Klaus Maier-Hein, MEYER Philippe, Xavier Coubez, Sylvain Faisan

Keywords: PDAC, CT segmentation, uncertainty, calibration

TL;DR: A lightweight post-hoc method that calibrates segmentation probabilities to the mean human response, making predictions more interpretable under multi-rater ambiguity.

Registration Requirement: Yes

Abstract:

Pancreatic ductal adenocarcinoma (PDAC) segmentation on contrast-enhanced CT is inherently ambiguous: inter-rater disagreement among experts reflects genuine uncertainty rather than annotation noise. Standard deep learning approaches assume a single ground truth, producing probabilistic outputs that can be poorly calibrated and difficult to interpret under such ambiguity. We present TwinTrack, a framework that addresses this gap through post-hoc calibration of ensemble segmentation probabilities to the empirical mean human response (MHR) -- the fraction of expert annotators labeling a voxel as tumor.
Calibrated probabilities are thus directly interpretable as the expected proportion of annotators assigning the tumor label, explicitly modeling inter-rater disagreement.
The proposed post-hoc calibration procedure is simple and requires only a small multi-rater calibration set. It consistently improves calibration metrics over standard approaches when evaluated on the MICCAI 2025 CURVAS--PDACVI multi-rater benchmark.

Visa & Travel: No

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Submission Number: 11