Unlocking Telemetry Potential: Self-Supervised Learning for Continuous Clinical Electrocardiogram Monitoring
Applying machine learning to monitoring hospitalised patients has the potential to save lives. In this paper we take first steps toward that goal by demonstrating how deep learning can be successfully applied to the telemetry monitoring systems, in particular the electrocardiogram (ECG), which is ubiquitous in the modern hospital setting. By leveraging self-supervised learning we pretrain a series models across the large yet unlabeled telemetry dataset, and demonstrate improved model performance across many downstream tasks. We also provide qualitative demonstrations of how these models an effectively annotate ECG in real time based on scarce high quality labels, showing one path to AI-based hospital monitoring.
Large pretrained models outperform their supervised counterparts, and can effectively label ECG with clinically valuable information