CAMEL: An ECG Language Model for Forecasting Cardiac Events.

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Electrocardiograms (ECG) are vital for diagnosing heart conditions.
• ECG language models (ELMs) show promise for ECG classification and report generation.
• Current ELMs lack the ability to predict future cardiac events.

Purpose of the Study:

• To develop the first ECG language model (ELM) with forecasting capabilities.
• To enable prediction of future cardiac events for proactive intervention.
• To introduce CAMEL, an ELM designed for longer ECG signal analysis.

Main Methods:

• Developed CAMEL, an ELM featuring a specialized ECG encoder for cross-modal signal-text understanding.
• Employed LoRA adaptation and a curriculum learning pipeline for training.
• Included ECG classification, metrics calculation, and conversational reasoning in the training curriculum.

Main Results:

• CAMEL achieved strong zero-shot performance across 6 tasks and 9 datasets.
• Introduced ECGForecastBench, a new benchmark for arrhythmia forecasting.
• CAMEL outperformed existing ELMs and fully supervised models on ECGBench and ECGForecastBench.

Conclusions:

• CAMEL is the first ELM capable of forecasting future cardiac events.
• The specialized ECG encoder facilitates cross-understanding of ECG signals and text.
• CAMEL sets a new state-of-the-art in ECG analysis and prediction.