Fig. 5: Comprehensive evaluation of ultraFedFM demonstrates high performance, strong generalization, and improved fairness.
From: From pretraining to privacy: federated ultrasound foundation model with self-supervised learning
a The quantitative comparison between UltraFedFM and state-of-the-art ultrasound task-specific models across four typical ultrasound tasks. b Generalization performance evaluation on out-of-distribution organ (i.e., kidney) and modality (i.e., high-frequency ultrasound). c The receiver operating characteristic curve (ROC) and precision-recall curve (PRC) on a new organ and new ultrasound imaging modality. d, e The quantitative analysis of UltraFedFM with state-of-the-art federated learning methods in terms of pre-training convergence (d) and prediction fairness (e).
