Establishment and validation of an artificial intelligence-based system for identifying the culprit vessel in...

The inaccurate identification of the culprit vessel is a critical issue in the treatment of patients with ST-segment elevated myocardial infarction (STEMI). This study aimed to establish an artificial intelligence (AI)-based algorithm for the culprit vessel identification in patients with STEMI.

698 ECGs were conducted within 12 hours before primary percutaneous coronary intervention from patients with STEMI. The internal training and testing sets consisted of 543 ECGs collected from two tertiary hospitals, and the external validation set consisted of 155 ECGs collected from another independent tertiary hospital. These ECGs represented STEMI patients with various culprit vessels, such as left anterior descending artery (LAD), right coronary artery (RCA) and left circumflex artery (LCX), validated by further coronary angiography (CAG).

Among 698 patients (mean (SD) age, 71.1 (11.3) years), the culprit vessels were LAD in 328, RCA in 288 and LCX in 82. In the internal training, an algorithm was developed (LAD: sensitivity 92.4%, specificity 99.7%; RCA: sensitivity 93.2%, specificity 97.4%; LCX: sensitivity 99.7%, specificity 95.8%). Internal testing revealed that it achieved high performance metrics (LAD: sensitivity 91.6%, specificity 96.0%; RCA: sensitivity 75.1%, specificity 95.8%; LCX: sensitivity 97.0%, specificity 88.8%), outperforming cardiologists and commercial algorithm. In the external validation, it also demonstrated competitive performance (LAD: sensitivity 72.0%, specificity 94.3%; RCA: sensitivity 90.5%, specificity 92.4%; LCX: sensitivity 92.9%, specificity 91.2%).

We established and validated an AI-based algorithm for the culprit vessel identification in patients with STEMI, comparable to experienced cardiologists. The generalisability of our algorithm on a global scale needs further verification.

NCT03317691.