Related Experiment Video
Updated: May 23, 2026
In Silico Clinical Trials for Cardiovascular Disease
Published on: May 27, 2022
2 Parallel Heart: Parallel Experiments of a Highly Accurate Dual Independent Neural Network for Predicting Myocardial
Objective:
To enhance acute myocardial infarction (AMI) prediction accuracy and interpretability by developing a dual-model parallel framework integrating electrocardiography (ECG) signals and blood biochemical markers.
Methods:
A convolutional neural network (CNN) extracted ECG features, while a backpropagation (BP) network analyzed 13 biochemical markers (troponin I/T, CK-MB, NT-proBNP, glucose, and the standard chemistry panel). Models were trained and validated on the PTB-XL ECG dataset ($n=21{,}799$), the MIMIC-III biochemical database ($n=4{,}760$), and a Chinese ECG dataset ($n=503$) for regional calibration.
Results:
The CNN-based ECG model achieved an AUC of 0.933, with P-R interval and QRS duration as significant predictors. The BP-based biochemical model achieved an AUC of 0.960. The integrated parallel framework achieved a total accuracy of 96.6%, significantly reducing missed diagnoses for non-ST-segment elevation myocardial infarction (NSTEMI) patients with inconspicuous ECG changes.
Conclusion:
This dual-model framework provides an efficient automated screening tool for AMI, bridging ECG and biochemical diagnostics and highlighting the importance of ethnic-specific calibration.
Significance:
By recovering NSTEMI cases that ECG-only models systematically miss and by showing that a regional cohort can recalibrate a European-trained classifier, the framework extends automated AMI screening to the presentations and populations where it currently fails most often.