Fully Automated Diagnosis of Acute Myocardial Infarction Using Electrocardiograms and Multimodal Deep Learning

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Rapid detection of acute myocardial infarction (AMI) is critical for reducing patient morbidity and mortality.
• Deep learning (DL) offers potential for enhancing automated interpretation of electrocardiogram (ECG) data.

Purpose of the Study:

• To develop and validate a DL model for AMI detection.
• The model integrates ECG data, patient demographics, and reported symptoms.

Main Methods:

• A retrospective cohort study utilized ECG data from 104,507 individuals across three Swedish centers.
• A residual convolutional neural network was trained on ECG features, age, sex, and symptoms.
• Model performance was evaluated using area under the receiver operating characteristic (AUROC), sensitivity, and specificity.

Main Results:

• The DL model achieved an AUROC of 0.8221 ± 0.0101 for internal validation and 0.8314 ± 0.0085 for external validation.
• Performance remained consistent across different sexes but was slightly lower for ambulance-arriving patients.
• Saliency maps indicated the model focused on ST segments and T waves for predictions.

Conclusions:

• The developed DL model demonstrates robust performance in detecting AMI across diverse patient populations.
• Further research, including a randomized trial, is recommended to compare the model's efficacy against emergency physician interpretations.