Authors: Sang-Hyup Lee, Kyu Lee Jeon, Yong-Joon Lee, Seng Chan You, Seung-Jun Lee, Sung-Jin Hong, Chul-Min Ahn, Jung-Sun Kim, Byeong-Keuk Kim, Young-Guk Ko, Donghoon Choi, Myeong-Ki Hong
Published in: Annals of Emergency Medicine, July 26, 2024
Conclusion: The deep ensemble AI model exhibited outstanding and balanced performance in detecting STEMI. The model's explainability, visualized through gradient-weighted class activation mapping, was deemed reasonable. Further studies are needed to prospectively validate the model's clinical benefits in real-world settings.
Methods: The study utilized electrocardiography (ECG) waveform data from a prospective percutaneous coronary intervention registry in Korea. Two board-certified cardiologists independently established a criterion standard for each ECG as either STEMI or Not STEMI, using corresponding coronary angiography data. The researchers developed a deep ensemble model by integrating five convolutional neural networks. The AI model underwent clinical validation using a symptom-based ECG data set and was compared against the performance of clinical physicians, with additional external validation performed.
Results: The data set for model development included 18,697 ECGs, with 1,745 (9.3%) identified as STEMI. The AI model demonstrated an accuracy of 92.1%, a sensitivity of 95.4%, and a specificity of 91.8%. The model’s performance was robust and consistent across clinical validation, comparisons with clinical physicians, and external validation.
Lee, S.H., Jeon, K.L., Lee, Y.J., You, S.C., Lee, S.J., Hong, S.J., Ahn, C.M., Kim, J.S., Kim, B.K., Ko, Y.G. and Choi, D., 2024. Development of Clinically Validated Artificial Intelligence Model for Detecting ST-segment Elevation Myocardial Infarction. Annals of Emergency Medicine.