When Does Synthetic CT Transfer? A Label-Free Donor/Host Diagnostic for Medical Vision-Language Model Routing on Real Lung CT

2026-06-28

AI Summary

This study addresses the challenge of evaluating model transferability from synthetic to real medical images in the absence of ground-truth labels. The authors propose a label-free transferability prediction mechanism that leverages a digital twin of lung CT scans to disentangle the contributions of nodules (donors) and surrounding anatomical structures (hosts) to model decisions. By integrating vision-language models (VLMs), a training-free routing diagnostic framework termed TrialCouncil, and a leave-one-source-out calibration strategy, they analyze transfer behavior across three tasks, five VLMs, four prompting conditions, and seven real-world datasets. Results reveal that nodule presence and size ranking are highly transferable (R² ≥ 0.96), whereas lobe localization is not. TrialCouncil accurately identifies transferable regions, matching the performance of the best fixed model. This work is the first to demonstrate the decisive role of donor/host-driven mechanisms in determining transfer success.

Abstract

A synthetic measurement of model competence is useful only if it survives the move to real data, yet the real labels that would verify it are exactly what medical imaging lacks. We ask whether transfer can be predicted in advance, label-free, and answer with a mechanism: on synthetic digital twins, competence that is donor-driven (a property of the transplanted nodule) survives the synthetic to real change of host, while host-driven competence (a property of the surrounding anatomy) need not. We test this on three lung CT vision-language tasks chosen to span that axis, across five public VLMs, four guidance conditions, and seven real datasets. The prediction holds in every case: presence and size orderings transfer (R2 >= 0.96), lobe does not; the split survives leave-source-out calibration, and the diagnostic names that boundary before any real label. TrialCouncil, a training-free council calibrated only on synthetic CT, confirms it by matching the best fixed model exactly where transfer is predicted. The contribution is not the router but the finding that transfer itself is predictable, label-free, from synthetic data alone.

Problem

• synthetic CT
• transfer prediction
• label-free
• medical vision-language model
• model competence

Innovation

• synthetic CT
• label-free transfer
• vision-language model
• donor/host competence
• model routing