Early detection of acute myocardial infarction (AMI) is essential, as treatment delay substantially reduces the effectiveness of reperfusion therapy and increases mortality. However, the time from symptom onset to first medical contact is typically at least 2 hours, largely due to nonspecific symptoms and the lack of diagnostic tools outside clinical settings. Although smartwatches enable ECG acquisition at home, they provide only single- or two-lead recordings and cannot capture precordial leads required for reliable AMI diagnosis. Furthermore, deep learning (DL) models for AMI detection often fail to generalize under dataset shift and may rely on clinically irrelevant features.
The AMI-Sure project addresses these limitations through an integrated hardware-algorithm system for diagnostic-grade ECG acquisition and trustworthy AMI detection in home environments. A chest accessory compatible with a wrist-worn device will enable near-standard 12-lead ECG acquisition with a single touch. A person-specific model will then be applied to reconstruct a diagnostically valid 12-lead ECG from a near-standard ECG. In parallel, a transfer learning-based AMI detector will be developed and systematically evaluated under covariate and temporal dataset shifts, with interpretability integrated to promote physiologically consistent decision-making. Clinical utility will be assessed in a prospective study in which patients will self-acquire ECGs on different days. Validation will include comparison with standard 12-lead ECGs, blinded cardiologist evaluation, and quantitative assessment of automated AMI detection performance.
The Lithuanian and Taiwanese teams bring expertise in wearable hardware development, conductive textiles, person-specific ECG reconstruction, DL, and clinical validation. By enabling early AMI detection at home, AMI-Sure has the potential to reduce treatment delays, improve access to care, and contribute to the digital transformation of cardiovascular medicine.
Project funding:
Interngovernmental programme administrated by Research Council of Lithuania: Lithuania– Taiwan
Project results:
Main project results will include a patentable ECG acquisition and reconstruction approach, a validated AMI progression simulator, an annotated clinical wrist–chest ECG database, and DL-based AMI detection methods supported by interpretability analysis. Results will be disseminated through high-level scientific publications, an international conference, open-source software, and publicly accessible research data.
Period of project implementation: 2026-10-01 - 2028-09-30
Project coordinator: Kaunas University of Technology
Project partners: Vilnius University
