Beatrice Zanchi

Beatrice Zanchi, Giuliana Monachino, Alvise Dei Rossi et al.

Background: Foundation models (FMs) trained on large-scale unlabeled physiological data have emerged as a promising paradigm for medical artificial intelligence. Their ability to capture clinically meaningful, transferable representations for rare diseases remains largely unproven. This study investigates whether FM pre-training provides genuine clinical generalization benefits beyond improved optimization for rare electrocardiographic (ECG) phenotypes. Methods: We systematically evaluated nine publicly available ECG FMs for Brugada syndrome detection on the BrSwiss cohort (294 patients, 87 cases) and the independent external HUCA cohort (363 patients, 76 cases), under three strategies (from-scratch training, linear probing, full fine-tuning) across several configurations, including a 3% data ablation and zero-shot cross-site transfers. Results: Pre-training was necessary for high-capacity architectures unable to converge from scratch (AUC gain up to 0.411, p < 0.05), but gave no significant gain for compact architectures already converged on labeled data alone. On full BrSwiss, the best fine-tuned FM (ECG-CPC, AUC = 0.962) only marginally exceeded the strongest supervised baseline (ECG-CPC from scratch, AUC = 0.932; p = 0.091). At matched training-set size, the data-efficiency advantage on BrSwiss-3% (AUC gain = 0.055, p < 0.01) did not replicate on HUCA. Under zero-shot cross-site transfer, FM-based pipelines did not generalize better than supervised baselines, all approaching chance-level performance. Conclusion: For Brugada syndrome detection, FM pre-training is mechanical rather than semantic, providing optimization stability rather than transferable clinical knowledge. These findings challenge the assumption that large-scale pre-training inherently encodes clinically meaningful representations, highlighting the central role of model architecture and data-domain alignment.

Giuliana Monachino, Nicolò La Porta, Beatrice Zanchi et al.

Foundation Models (FMs) are large-scale machine learning models trained on extensive, diverse datasets that can be adapted to a wide range of downstream tasks with minimal fine-tuning. In the last two years, interest in FMs has also grown for applications in the cardiological field to analyze the electrocardiogram (ECG) signals. One of the key properties of FMs is their transferability to a wide range of downstream scenarios. With the spread of wearable and portable devices, keen interest in learning from reduced-channel configurations has arisen. However, the adaptation of ECG FMs to downstream scenarios with fewer available channels still has to be properly investigated. In this work, we propose Self-DANA, a novel, easy-to-integrate solution that makes self-supervised architectures adaptable to a reduced number of input channels, ensuring resource efficiency and high performance. We also introduce Random Lead Selection, a novel augmentation technique to pre-train models in a more robust and channel-agnostic way. Our experimental results on five reduced-channel configurations demonstrate that Self-DANA significantly enhances resource efficiency while reaching state-of-the-art performance. It requires up to 69.3% less peak CPU memory, 34.4% less peak GPU memory, about 17% less average epoch CPU time, and about 24% less average epoch GPU time.