Empirical investigation of 3D CT Foundation Models and Unsupervised Adaptation for Head and Neck Cancer Recurrence Prediction

2026-07-29

AI Summary

This study systematically evaluates the cross-institutional generalization capability of 3D CT foundation models for predicting head and neck cancer recurrence and investigates the necessity of downstream task adaptation. Leveraging two public datasets comprising 3,644 patients, we present the first large-scale empirical analysis of multiple 3D CT foundation models under real-world clinical distribution shifts, comparing the effectiveness of unsupervised domain adaptation strategies against multimodal fusion mechanisms. Our results demonstrate a significant performance drop in external validation, highlighting the substantial challenge of cross-center generalization. Notably, approaches integrating imaging with clinical data consistently achieve the best predictive performance, confirming that multimodal integration remains the most effective paradigm for this task.

Abstract

The rapid emergence of 3D CT foundation models has opened new avenues for predictive modeling from CT imaging, offering a compelling alternative to traditional radiomics which is known to suffer from reproducibility issues and sensitivity to acquisition protocol variations. Yet, as these models grow in availability, a critical need arises to evaluate how well their learned representations generalize across diverse clinical settings and whether adaptation to specific downstream tasks is necessary to unlock their full potential. To address these questions, we benchmarked several 3D CT foundation models for predicting recurrence-free survival in head and neck cancer across two public datasets totaling 3,644 patients, evaluating various adaptation strategies and modality fusion mechanisms. Our findings reveal persistent difficulty in identifying features that generalize consistently across different imaging distributions, as evidenced by significant performance drops on external validation cohorts. Ultimately, the integration of imaging features with clinical data remains the most accurate approach for prognostic prediction, though achieving universal generalization across varied clinical contexts continues to represent a substantial challenge for the current generation of models.

Problem

• 3D CT foundation models
• unsupervised adaptation
• head and neck cancer
• recurrence prediction
• generalization

Innovation

• 3D CT foundation models
• unsupervised adaptation
• head and neck cancer
• recurrence prediction
• cross-domain generalization