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

arXiv: https://arxiv.org/abs/2606.29232 · HTML: https://mlsanity.com/paper/2606.29232/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Medical vision-language models, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 22 of its 30 references that are in the map.
- First posted on arXiv 28 Jun 2026.

## Placed in

- Multimodal LLMs / vision-language models / [Medical vision-language models](https://mlsanity.com/mllm/medical-vision-language-models/)

## Builds on (most cited first, 22 of 22)

- [Self-Consistency Improves Chain of Thought Reasoning in Language Models](https://mlsanity.com/paper/2203.11171/) (2022)
- [Qwen2.5-VL Technical Report](https://mlsanity.com/paper/2502.13923/) (2025)
- [Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer](https://mlsanity.com/paper/1701.06538/) (2017)
- [Shortcut Learning in Deep Neural Networks](https://mlsanity.com/paper/2004.07780/) (2020)
- [LLaVA-Med: Training a Large Language-and-Vision Assistant for Biomedicine in One Day](https://mlsanity.com/paper/2306.00890/) (2023)
- [Selective Classification for Deep Neural Networks](https://mlsanity.com/paper/1705.08500/) (2017)
- [BiomedCLIP: a multimodal biomedical foundation model pretrained from fifteen million scientific image-text pairs](https://mlsanity.com/paper/2303.00915/) (2023)
- [MedGemma Technical Report](https://mlsanity.com/paper/2507.05201/) (2025)
- [Consistent Estimators for Learning to Defer to an Expert](https://mlsanity.com/paper/2006.01862/) (2020)
- [Lingshu: A Generalist Foundation Model for Unified Multimodal Medical Understanding and Reasoning](https://mlsanity.com/paper/2506.07044/) (2025)
- [Validation, comparison, and combination of algorithms for automatic detection of pulmonary nodules in computed tomography images: the LUNA16 challenge](https://mlsanity.com/paper/1612.08012/) (2016)
- [Vision-Language Models for Medical Report Generation and Visual Question Answering: A Review](https://mlsanity.com/paper/2403.02469/) (2024)
- [MAISI: Medical AI for Synthetic Imaging](https://mlsanity.com/paper/2409.11169/) (2024)
- [CARES: A Comprehensive Benchmark of Trustworthiness in Medical Vision Language Models](https://mlsanity.com/paper/2406.06007/) (2024)
- [LNDb: A Lung Nodule Database on Computed Tomography](https://mlsanity.com/paper/1911.08434/) (2019)
- [Open (Clinical) LLMs are Sensitive to Instruction Phrasings](https://mlsanity.com/paper/2407.09429/) (2024)
- [To Ensemble or Not: Assessing Majority Voting Strategies for Phishing Detection with Large Language Models](https://mlsanity.com/paper/2412.00166/) (2024)
- [Regression with Cost-based Rejection](https://mlsanity.com/paper/2311.04550/) (2023)
- [Virtual Lung Screening Trial (VLST): An In Silico Study Inspired by the National Lung Screening Trial for Lung Cancer Detection](https://mlsanity.com/paper/2404.11221/) (2024)
- [The Utility of the Virtual Imaging Trials Methodology for Objective Characterization of AI Systems and Training Data](https://mlsanity.com/paper/2308.09730/) (2023)
- [NodMAISI: Nodule-Oriented Medical AI for Synthetic Imaging](https://mlsanity.com/paper/2512.18038/) (2025)
- [iTRIALSPACE: Programmable Virtual Lesion Trials for Controlled Evaluation of Lung CT Models](https://mlsanity.com/paper/2605.05761/) (2026)

## Abstract (arXiv)

Many medical vision--language models (VLMs) are publicly available, but a site adopting one for a new imaging task rarely has the labeled real data needed to compare them. Because synthetic CT has ground truth by construction, a site could instead rank candidate models on synthetic scans, if the ranking carries over to real data. We study when it does for lung CT, using five public VLMs and three routine screening tasks, namely detecting a nodule, locating its lobe, and estimating its size, on 85,021 synthetic and 26,085 real slices from seven datasets. For nodule presence and size, the synthetic ranking matches the real one with a rank correlation of 0.98. For lobe, the ranking is only moderately preserved, with a rank correlation of 0.80. We also propose TrialCouncil, a training-free method that uses synthetic accuracy to select one model, combine several, or defer uncertain cases. On real CT, it outperforms standard voting methods in every lobe and size setting and outperforms the best synthetic model on lobe, while selecting the best synthetic model works better for presence. Synthetic CT can therefore serve as a first safeguard when choosing a medical VLM without labeled real data.