Key result
A multimodal deep learning model integrating 10-second ECG waveforms and patient demographics achieved a mean AUROC of 92.1% for predicting myocardial infarction, significantly outperforming ECG-only models.
Why the study?
Existing ML models for MI detection largely distinguish MI from healthy controls without accounting for confounding conditions, and fail to leverage clinical data beyond the 12-lead ECG.
Does a multimodal machine learning model integrating 10-second ECG waveforms and patient demographics improve the prediction of acute myocardial infarction compared to ECG-only models?
Does a multimodal machine learning model integrating 10-second ECG waveforms and patient demographics improve the prediction of acute myocardial infarction compared to ECG-only models?
Absolute Event Rate: 92.1% vs 91.5%
p-value: p=<0.001
Integrating 10-second ECG waveforms and patient demographics into a deep learning model significantly improves the diagnostic accuracy for early detection of myocardial infarction compared to standard 2.5-second ECG-only models.
Integrating demographics with ECGs may marginally improve early MI detection; leaves open whether this AI approach translates to.
Early identification and recognization of myocardial ischemia/infarction (MI) is the most important goal in the management of acute coronary syndrome (ACS). The 12-lead electrocardiogram (ECG) is widely used as the initial screening test for patients with chest pain but its diagnostic accuracy remains limited. There is an ongoing effort to address the issue with machine learning (ML) algorithms which have demonstrated improved performance. Most studies are designed to classify MI from healthy controls and thus are limited due to the lack of consideration of potential confounding conditions in the setting of MI. Moreover, other clinical information in addition to ECG has not yet been well leveraged in existing machine learning models. The present study aims to advance ML-based prediction models closer to clinical applications for early MI detection. The study considered downstream clinical implementation scenarios in the initial model design by dichotomizing study samples into MI and non-MI classes. Two separate experiments were then conducted to systematically investigate the impact of two important factors entrained in the modeling process, including the duration of ECG (2.5s vs. 10s), and the value of multimodal information for model training. A novel feature-fusion deep learning architecture was proposed to learn joint features from both ECG and patient demographics as the additional data modality. The best-performing model achieved a mean area under the receiver operating characteristic curve (AUROC) of 92.1% and a mean accuracy of 87.4%, which is on par with existing studies despite the increased task difficulty due to the new class design. The results also show that the ML model can capitalize on the information added from both the extra ECG waveforms in time and patient demographics. The findings in this study help guide the development of machine learning solutions for early MI detection and move the models one step closer to real-world clinical applications.
No takes yet. Share an insight, caveat, or question.
Xiao et al. (2022) studied Acute Myocardial Infarction (n=18,885). Multimodal deep learning model (10s ECG + patient demographics) vs. ECG-only deep learning model was evaluated on Mean Area Under the Receiver Operating Characteristic Curve (AUROC) for predicting myocardial infarction (p=<0.001). A multimodal deep learning model integrating 10-second ECG waveforms and patient demographics achieved a mean AUROC of 92.1% for predicting myocardial infarction, significantly outperforming ECG-only models.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: