WorkResearch Engineer · University of Glasgow · Apr 2025 — Present
Enhancing the Glasgow ECG Analysis Program
心電図
A deep-learning layer that improves atrial-fibrillation detection in the Glasgow ECG program.
The Glasgow ECG analysis program has been developed for more than fifty years in the ECG Core Lab at the University of Glasgow, led by Prof. Peter Macfarlane and Prof. Derek Connelly. It helps read over 20 million recordings a year across licensed devices and research studies. My main project is a commercial deep-learning layer that improves how the program detects atrial fibrillation. It strengthens the existing algorithm instead of replacing it, which cuts false positives and lets clinicians set the balance of sensitivity and specificity they want.
What I do
- Building the atrial-fibrillation detection models with task-specific machine learning. I chose that over a large foundation model to keep the operating point interpretable and under our control.
- Reading the atrial residual of the ECG with weak-signal methods borrowed from astronomy and geophysics (Lomb-Scargle periodograms, chaos and entropy measures, Allan and Fano variance).
- Validating against expert-adjudicated reference standards reviewed by Prof. Macfarlane and Prof. Connelly, across external databases including MIMIC-IV, PTB-XL, and CODE-15.
- Bringing rigorous evaluation, testing, and traceability to a long-lived, safety-critical codebase.