Bionic Wearable ECG with Multimodal Large Language Models: Coherent Temporal Modeling for Early Ischemia Warning and Reperfusion Risk Stratification - PubMed
4 hours ago
- #AI in healthcare
- #Wearable technology
- #Biomedical monitoring
- Proposes a framework combining bionic wearable ECG sensors and multimodal large language models for early ischemia warning and reperfusion risk stratification.
- Utilizes a temporally hierarchical fusion transformer with cross-granularity attention to model intrabeat, interbeat, and long-term dependencies simultaneously.
- Validated on four datasets with 108,778 patients, achieving an AUROC of 0.947 for ischemia detection and a C-index of 0.923 for risk stratification, outperforming baselines by 4.8% to 9.5%.
- Provides an average lead time of 18.4 minutes before ischemic events, enabling timely clinical interventions.