Adapting Medical Foundation Models for Coronary Artery Calcium Segmentation from CT Imaging

Jie-En Tsai, Po-Chih Kuo

Keywords: Foundation Model, Segmentation, Coronary Artery Calcium, Fairness

Registration Requirement: Yes

Abstract: Automated coronary artery calcium (CAC) segmentation plays an important role in coronary artery disease risk stratification; however, conventional task-specific deep learning models typically require large annotated datasets that are challenging to obtain in clinical settings. In this study, we address this limitation by fine-tuning a medical imaging foundation model for CAC segmentation and evaluating its performance under progressively reduced training data. The proposed approach outperforms existing methods when trained on the full dataset and remains competitive even with only 50% of the training data. Moreover, it exhibits smaller performance disparities across sex, age, and CAC severity subgroups compared to a U-Net baseline, highlighting the potential of foundation model fine-tuning for robust and equitable clinical AI applications.

Visa & Travel: No

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Submission Number: 19