Development of novel machine learning model for right ventricular quantification on echocardiography-A multimodality...

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Echocardiography (echo) is crucial for right ventricular (RV) assessment.
• Current RV evaluation methods are time-consuming, requiring manual analysis.
• Machine learning (ML) offers potential for automated RV function quantification.

Purpose of the Study:

• To develop and validate an automated ML model for RV assessment.
• To quantify RV function through automated tricuspid annulus tracking.
• To compare ML-derived indices with cardiac magnetic resonance (CMR) reference standards.

Main Methods:

• Developed a convolutional neural network (CNN) model for automated tricuspid annulus tracking on echo.
• Trained the model on 7791 image frames.
• Generated automated linear and circumferential indices of annular displacement.
• Compared automated indices to CMR-defined RV dysfunction (RVEF < 50%).

Main Results:

• Fully automated annular tracking was successful in 101 patients with minimal processing time (<1 second).
• Automated annular shortening indices were significantly lower in patients with RV dysfunction (P < .001).
• ML-derived indices showed good diagnostic performance (AUC 0.69-0.73) and high NPV (84%-87%).
• ML algorithm demonstrated superior reproducibility (ICC 1.0) compared to manual segmentation (ICC 0.87-0.91).

Conclusions:

• Initial validation of a deep learning system for RV assessment using automated tricuspid annulus tracking.
• The ML approach provides efficient, reproducible, and accurate RV function quantification.
• This automated method has the potential to enhance clinical RV evaluation.