AutoRV: Automated Assessment of Right Ventricular Structure and Function Using Transesophageal Echocardiography in...

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Right ventricle (RV) dysfunction significantly impacts mechanically ventilated intensive care unit (ICU) patients.
• Current manual assessment of RV function is qualitative, time-consuming, and imprecise.
• Automated quantitative assessment is needed for effective RV monitoring.

Purpose of the Study:

• To develop and validate AutoRV, an automated pipeline for assessing RV function.
• To utilize deep learning for landmark detection in 2-D transesophageal echocardiography (TEE).
• To enable precise and efficient quantitative RV function analysis.

Main Methods:

• Implemented deep convolutional neural networks (CNNs) for automatic RV apex and tricuspid annulus landmark detection in 2-D TEE images.
• Trained and tested networks on data from 51 patients.
• Computed key RV function parameters: tricuspid annular plane systolic excursion, RV fractional area change, and linear RV strain (RVS_global).
• Compared automated results against manual measurements by an experienced clinician.

Main Results:

• The U-Net architecture achieved high performance, processing images at 228 frames/s.
• Landmark detection accuracy was 4.2±3.4mm from ground truth.
• Parameter errors were: tricuspid annular plane systolic excursion (1.1±2.3mm), RV fractional area change (0.2±6.6%), and RVS_global (3.9±3.6%).
• Agreement between AutoRV and manual measurements matched inter-observer variability.

Conclusions:

• AutoRV is a fully automatic, end-to-end solution for RV function assessment.
• The system demonstrates acceptable accuracy and significant time efficiency.
• AutoRV shows potential for routine RV monitoring in ICU patients.