Generalization of the T2w VERIDAH model across scanners

Determine whether the VERIDAH model trained on T2-weighted TSE MRI from the NAKO cohort achieves comparable performance on T2-weighted TSE MRI acquired on different scanners.

Background

The VERIDAH T2w model was trained exclusively on the NAKO cohort. While the CT dataset used in the study is diverse, the T2w MRI training data come from a single source, leading to uncertainty about how the model performs on data acquired with different scanner hardware or protocols.

Assessing cross-scanner generalization is important for clinical deployment where acquisition settings vary widely.

References

Since the VERIDAH model trained on T2w was only trained on the NAKO cohort, we cannot guarantee the same performance on T2w TSE MRI from a different scanner.

— VERIDAH: Solving Enumeration Anomaly Aware Vertebra Labeling across Imaging Sequences  (2601.14066 - Möller et al., 20 Jan 2026) in Section 6 (Discussion and Conclusion)