DiffuSETS: 12-Lead ECG generation conditioned on clinical text reports and patient-specific information

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiology

Background:

• High-quality electrocardiogram (ECG) data is scarce due to privacy and resource limitations.
• Existing ECG generation methods often use small datasets and lack robust evaluation.
• There's a need for advanced ECG signal generation for clinical applications.

Purpose of the Study:

• To introduce DiffuSETS, a framework for generating semantically aligned and high-fidelity ECG signals.
• To enable ECG generation from diverse clinical text reports and patient-specific data.
• To establish a comprehensive benchmarking methodology for evaluating ECG generative models.

Main Methods:

• Developed DiffuSETS, a novel framework for ECG signal synthesis.
• Utilized clinical text reports and patient data as input modalities.
• Implemented a rigorous benchmarking methodology for performance assessment.

Main Results:

• DiffuSETS demonstrated superior performance in generating clinically meaningful ECG signals.
• The framework achieved high semantic alignment and signal fidelity.
• Experimental results confirmed the model's effectiveness in ECG generation tasks.

Conclusions:

• DiffuSETS effectively addresses the challenge of ECG data scarcity.
• The framework shows potential for applications in cardiology education and medical knowledge discovery.
• This work advances the field of AI-driven medical signal generation.