Abstract—Echocardiography plays a vital role in assessing cardiac function, with left ventricular Ejection Fraction (EF) serving as a key indicator of systolic performance. However, privacy regulations and restrictions on sharing patient data across institutions make it difficult to develop robust machine learning models for EF assessment. Federated Learning (FL) offers a promising solution by enabling collaborative model training without transferring raw clinical data. This study presents a privacy-preserving FL framework for echocardiographic EF classification and evaluates its performance using the Cardiac Acquisitions for Multi-structure Ultrasound Segmentation (CAMUS) dataset under both Independent and Identically Distributed (IID) and Non-Independent and Identically Distributed (Non-IID) client data settings. In addition, the effect of injecting Gaussian noise into client updates is investigated as a lightweight privacy-enhancing strategy. For the clinically realistic EF classification task (EF < 50% versus EF ≥ 50%), centralised learning achieved an AUC of 0.765, whereas FL under IID and Non-IID settings achieved AUCs of 0.746 and 0.702, respectively. For the extreme EF cohort classification task (EF ≤ 35% versus EF ≥ 60%), centralised learning and IID-FL achieved an AUC of 0.966, while Non-IID FL maintained the same Area under Curve (AUC) with a modest reduction in classification accuracy. The proposed framework also demonstrated strong resilience to moderate update perturbations, maintaining test AUC values above 0.96 across all evaluated noise levels. These findings highlight the potential of federated learning for privacy-aware, multi-institutional echocardiographic analysis and identify data heterogeneity as the primary challenge affecting federated model performance.
 
Keywords—federated learning, echocardiography, ejection fraction classification, privacy preservation, Non-Independent and Identically Distributed (Non-IID) data, medical image analysis
 
Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).