ECGEL: a multimodal 12-lead ECG classification model for heart failure prediction

Area of Science:

• Cardiology
• Artificial Intelligence in Medicine
• Biomedical Signal Processing

Background:

• Cardiovascular diseases (CVD) represent a leading global cause of mortality.
• Heart failure (HF) significantly diminishes patient quality of life and incurs substantial socioeconomic costs.
• Current HF diagnostic methods are often complex and costly, delaying critical early interventions.

Purpose of the Study:

• To develop an efficient and accurate multimodal model for early heart failure prediction.
• To address the limitations of current diagnostic approaches through data fusion.
• To improve patient outcomes by enabling timely intervention.

Main Methods:

• A multimodal model, ECGEL, was developed, integrating electrocardiogram (ECG) and clinical text data.
• ECG signals were denoised (LUNet), converted to spectrograms, and features extracted (EfficientNetv2).
• Clinical text data was preprocessed (Bert) and features extracted (BiLSTM), followed by feature fusion for prediction.

Main Results:

• The ECGEL model demonstrated high performance on a private dataset.
• Achieved an accuracy of 97.9%, recall of 98.3%, and F1 score of 97.6% for heart failure prediction.
• The multimodal approach proved effective in enhancing diagnostic accuracy.

Conclusions:

• ECGEL offers an efficient and accurate solution for early heart failure diagnosis.
• The model shows significant potential for clinical application in cardiovascular disease management.
• Integrating ECG and clinical text data improves the prediction of heart failure.