Fig. 2: Performance of individual OMAFound modules in cancer screening.
From: A foundation model for breast and lung cancer screening using non-contrast computed tomography
a–c, ROC curves of the CT-based OMAFound for breast cancer prediction (breast-specific module; a), lung cancer prediction (lung-specific module; b) and patient-level cancer prediction (fusion module; c). d–f, The feasibility of OMAFound for breast cancer screening compared with the standard mammography (MG) approach, assessed by the baseline of the mammography-based AI model (d), comparison between models on a paired CT-mammography dataset (e) and comparison on a subset of the paired CT–mammography dataset benchmarked against breast radiologists (f). All ROC curves are presented with a 95% confidence band.
