The MERL framework embodies a giant leap in bridging ECG representation learning with the wealth of clinical knowledge. MERL performs zero-shot classification through a multimodal approach, combining ECG data and informative text prompts. Clinical Knowledge Enhanced Prompt Engineering (CKEPE) uses LLMs to generate stronger prompts based on expert-verified data.

Key Highlights:

• Pioneers zero-shot multimodal learning for ECG classification.
• Eliminates the need for training data in downstream tasks.
• Uses expert-verified clinical knowledge base to enhance prompt quality.
• Outperforms existing self-supervised learning methods in benchmark tests.

MERL’s approach presents transformative potential for medical diagnostics, where accessibility to expert-level analysis could become widespread.