Development and validation of machine learning algorithms based on electrocardiograms for cardiovascular diagnoses at...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Electrocardiograms (ECGs) are crucial for cardiovascular (CV) diagnosis.
• AI is enhancing ECG interpretation for conditions beyond traditional measures.
• Early detection of diverse CV conditions remains a clinical challenge.

Purpose of the Study:

• To develop and validate AI models for the simultaneous prediction of 15 common CV diagnoses using ECGs.
• To assess the performance of deep learning (DL) and extreme gradient boosting (XGB) models.
• To evaluate AI's capability in population-level CV condition identification.

Main Methods:

• Retrospective study of 1,605,268 ECGs from 244,077 adult patients (Alberta, Canada, 2007-2020).
• Utilized ResNet-based deep learning (DL) on ECG tracings and XGB on ECG measurements.
• Included 15 International Classification of Diseases, 10th revision (ICD-10) CV diagnoses.

Main Results:

• DL models achieved Area Under the Receiver Operating Characteristic Curve (AUROC) >90% for 4 diagnoses (AVB, HCM, MS, STEMI).
• AUROC ranged from 80-90% for 8 conditions (CA, NSTEMI, VT, MVP, PHTN, AS, AF, HF).
• DL models outperformed XGB models by approximately 5% in AUROC.

Conclusions:

• AI-powered ECG analysis demonstrates good-to-excellent predictive performance for common cardiovascular conditions.
• DL models show significant potential for early and simultaneous detection of multiple CV diagnoses.
• These findings support the integration of AI in routine ECG interpretation for improved cardiovascular care.