Predicting post-operative right ventricular failure using video-based deep learning.

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Echocardiogram data offers rich temporal information but is often underutilized.
• Current echocardiography AI systems automate limited metrics, not full data patterns.
• Subjective assessments hinder clinical decisions, especially for predicting post-operative complications.

Purpose of the Study:

• To develop and evaluate a video AI system for predicting post-operative right ventricular failure (RV failure).
• To leverage the full spatiotemporal data within pre-operative echocardiograms.
• To improve clinical decision-making in patients undergoing mechanical circulatory support.

Main Methods:

• Trained a video AI system on pre-operative echocardiograms.
• The AI system utilized the complete spatiotemporal information from the echocardiograms.
• Evaluated the AI system's performance against human experts on independent data.

Main Results:

• Achieved an Area Under the Curve (AUC) of 0.729 in predicting RV failure.
• The AI system demonstrated significantly superior performance compared to a team of human experts.
• Successfully predicted post-operative RV failure using comprehensive echocardiogram analysis.

Conclusions:

• A video AI system can effectively predict post-operative RV failure from echocardiograms.
• AI systems utilizing full spatiotemporal data surpass traditional metrics and human expert performance.
• This approach holds promise for enhancing clinical decision-making in complex cardiac conditions.