Large Language Models in Intracardiac Electrogram Interpretation: A New Frontier in Cardiac Diagnostics for Pacemaker...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Intracardiac electrogram (EGM) interpretation demands specialized expertise often lacking among cardiologists.
• Advanced AI models, such as ChatGPT-4o, present a potential avenue for enhancing diagnostic accuracy in EGM analysis.
• This research systematically assesses ChatGPT-4o's capabilities in interpreting EGMs under varying levels of contextual information.

Purpose of the Study:

• To evaluate the diagnostic performance of ChatGPT-4o in interpreting intracardiac electrograms (EGMs).
• To assess the impact of increasing clinical context on ChatGPT-4o's EGM interpretation accuracy and reliability.
• To compare ChatGPT-4o's performance across different scenarios with varying complexity.

Main Methods:

• Twenty intracardiac electrogram (EGM) cases were analyzed using ChatGPT-4o.
• Performance was evaluated across four scenarios (A-D) with increasing contextual data.
• Statistical analysis included McNemar's test, Cohen's Kappa, and Prevalence- and Bias-Adjusted Kappa (PABAK) over two months.

Main Results:

• ChatGPT-4o's accuracy improved from 57% (Scenario A) to 66% (Scenario B) with added clinical context.
• Agreement was high for atrial activity and synchronization in Scenario A, but poor for chamber identification.
• Performance in complex scenarios (C and D) was lower, with fair consistency over time, indicating limitations in complex decision-making.

Conclusions:

• ChatGPT-4o demonstrates promising accuracy and reliability for structured EGM interpretation tasks.
• The AI model effectively integrates contextual information but shows limited adaptability to complex clinical cases.
• Further research and validation are necessary before widespread clinical adoption of AI in EGM interpretation.