Automated Assessment of Right Atrial Pressure From Ultrasound Videos Using Machine Learning.

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Early recognition of volume overload is critical for heart failure management.
• Ultrasound estimation of right atrial pressure (RAP) assesses intravascular volume status but requires expert interpretation.
• Limited availability of experienced physicians hinders widespread use of ultrasound-based RAP assessment.

Purpose of the Study:

• To evaluate the accuracy of machine learning models in estimating RAP from echocardiogram studies.
• To develop an automated deep learning approach for RAP estimation.
• To assess the robustness and generalizability of the developed algorithm.

Main Methods:

• Developed automated deep learning models to identify inferior vena cava scans and estimate RAP.
• Trained and evaluated models using 15,828 echocardiogram videos and 319 right heart catheterization measurements.
• Validated model performance against cardiologist estimates and external datasets.

Main Results:

• The model achieved 80.3% agreement with cardiologist RAP estimates (AUROC 0.844).
• RAP estimates were statistically indistinguishable from cardiologist and right heart catheterization measurements (P=0.98).
• The model demonstrated strong generalization to external echocardiogram data (AUROC 0.854).

Conclusions:

• Machine learning can accurately and robustly interpret RAP from echocardiogram videos.
• This automated algorithm facilitates objective assessments of intravascular volume status.
• The technology holds potential for widespread clinical application in heart failure management.