Authors: (including presenting author): :
Chan OY(1), Lui CT(1), Hung KL(1), Chan WC(1), Li CK(1), Tsang MY(1), Eric Kong(2), Kenny Kwok(2), Peter Chan(2), Edwin Lo(2)
Affiliation: :
(1) Department of Accident and Emergency, Tuen Mun Hospital (2) Information Technology Department, NTWC
Keyword 1: :
artificial intelligence
Keyword 2: :
ST-elevation myocardial infarction
Keyword 3: :
emergency department
Keyword 4: :
electrocardiography
Keyword 5: :
reprefusion time
Introduction: :
Early identification of ST-elevation myocardial infarction (STEMI) in A&E allows timely reperfusion and improve survival. In daily A&E workflow, triage nurses acquired fast-track ECGs for chest pain patients and emergency physicians interpreted them for early STEMI identification. Diagnostic errors in subtle STEMI are not uncommon worldwide, also traditional workflows failed to prioritize high-risk ECGs. Recent digital ECG transformation by Hospital Authority enables AI-assisted analysis to improve clinical care.
Objectives: :
To integrate artificial intelligence (AI) STEMI alert system into ECG screening workflow to expedite accurate STEMI identification and early reperfusion treatment.
Methodology: :
PMcardio STEMI AI ECG model (Powerful Medical), a neural-network deep learning model, was integrated into the ECG workflow in Tuen Mun A&E in February 2025 for non-emergent chest pain patients with triage-initiated ECGs. The setup rides on a Software-as-a-service (SAAS) design leverage on a designated edge tablet device intercalated between compatible ECG machines and CMS for paperless ECG. The architecture had been cleared on HA privacy and security assessment. The end-to-end analytic time of the service is less than 10 seconds. When STEMI ECGs were detected, “broken heart” alert notified supporting staff to alert doctors for immediate action.
Result & Outcome: :
From February to July 2025, 4153 AI ECG were performed with 82 STEMI alert. Referencing to the outcome of ED diagnosis of STEMI, AI detection demonstrated sensitivity 77% and specificity 98%, outperforming physician screening with 53% sensitivity and 99% specificity quoted from previous study. Mean door-to-clinical-escalation time improved from 29 in 2024 to 14 minutes for those with AI analysis performed in 2025, and median door-to-balloon time reduced from 103 in 2024 to 96 minutes. Incorporating AI into ECG screening improved STEMI detection accuracy and shortened time-critical reperfusion. In the future, AI ECG screening workflow can be explored to integrate in the chest pain green channel, and extension to cover prehospital ECG which would be further expediting the timely cardiac care.
