Unlocking the diagnostic potential of electrocardiograms through information transfer from cardiac magnetic resonance...

Area of Science:

• Artificial Intelligence in Medicine
• Cardiology
• Medical Imaging

Background:

• Electrocardiogram (ECG) is a cost-effective tool for cardiac functional assessment but has limitations in classifying and localizing cardiovascular diseases (CVD).
• Cardiac Magnetic Resonance (CMR) imaging offers detailed structural information for CVD diagnosis but is limited by scan time and cost.
• There is a need for cost-effective and comprehensive cardiac screening methods that leverage widely available data.

Purpose of the Study:

• To develop a deep learning strategy for cost-effective cardiac screening using solely ECG data.
• To transfer domain-specific information from CMR imaging to ECG representations for enhanced diagnostic capabilities.
• To demonstrate the utility and generalizability of the proposed method for CVD risk and cardiac phenotype prediction.

Main Methods:

• A deep learning approach combining multimodal contrastive learning and masked data modelling was employed.
• Domain-specific information was transferred from CMR imaging to ECG representations.
• Extensive experiments were conducted using data from 40,044 UK Biobank subjects.

Main Results:

• The method demonstrated significant utility and generalizability for subject-specific CVD risk prediction and cardiac phenotype prediction using only ECG data.
• The novel multimodal pre-training paradigm improved risk prediction performance by up to 12.19% and phenotype prediction by up to 27.59%.
• Qualitative analysis confirmed that learned ECG representations incorporated information from CMR image regions of interest.

Conclusions:

• Deep learning, integrating multimodal information, can enhance ECG analysis for comprehensive cardiac screening.
• The proposed method offers a cost-effective solution for CVD risk and cardiac phenotype prediction, overcoming limitations of traditional modalities.
• The publicly available pipeline facilitates further research and clinical application of advanced ECG-based cardiac diagnostics.