Artificial intelligence (ChatGPT) ready to evaluate ECG in real life? Not yet!

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Electrocardiograms (ECGs) are crucial for diagnosing cardiac conditions.
• The integration of Artificial Intelligence (AI) into medical diagnostics is rapidly advancing.
• There is a growing need for accessible diagnostic tools, especially in areas with limited specialist access.

Purpose of the Study:

• To evaluate the effectiveness and accuracy of ChatGPT-based AI models in interpreting ECGs.
• To compare the diagnostic performance of AI models against cardiologists.
• To explore the potential clinical utility of AI in ECG interpretation.

Main Methods:

• 107 ECG cases of varying difficulty were analyzed.
• Three ChatGPT-based AI models (GPT-ECGReader, GPT-ECGAnalyzer, GPT-ECGInterpreter) were used.
• AI model performance was compared to that of two cardiologists using statistical tests (chi-square, Fisher exact).

Main Results:

• Cardiologists achieved a significantly higher accuracy (92.52%) than AI models (57.94%-62.62%).
• AI models showed slightly better performance in female patients, but not statistically significant.
• Cardiologists outperformed AI across all ECG difficulty levels and in diagnosing arrhythmia and structural heart disease patterns.

Conclusions:

• ChatGPT-based AI models show promise for ECG interpretation but currently lack sufficient clinical reliability.
• Human oversight by a cardiologist is essential for current AI diagnostic tools.
• Further research is needed to enhance AI accuracy, particularly for complex ECG interpretations.