Multimodal Cardiovascular Disease Detection Using ECG Image and EHR.

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Electrocardiogram (ECG) signals are crucial for cardiovascular disease prediction.
• Clinical practice often stores ECG data as images, necessitating image-based analysis.
• Integrating diverse data sources can enhance predictive accuracy.

Purpose of the Study:

• To evaluate the performance of single-modal and multi-modal fusion methods for cardiovascular disease prediction.
• To compare the informativeness of ECG images versus electronic health record (EHR) features.
• To determine the optimal data fusion strategy using ECG and EHR data.

Main Methods:

• Sequential ECG data were transformed into image formats.
• Single-modal analysis was performed using ECG images and EHR features separately.
• Multi-modal fusion methods were employed to combine ECG image and EHR data.
• Predictive model performance was evaluated for all methods.

Main Results:

• Electronic health record (EHR) features demonstrated higher informativeness compared to ECG images for disease prediction.
• The multi-modal fusion method significantly outperformed all single-modal approaches.
• Combined ECG and EHR data yielded superior predictive accuracy.

Conclusions:

• EHR data are a more potent predictor of cardiovascular disease than ECG images alone.
• Multi-modal fusion of ECG and EHR data represents a promising strategy for improving cardiovascular disease risk prediction.
• Future research should focus on advanced fusion techniques to leverage comprehensive patient data.