Echocardiographic view and feature selection for the estimation of the response to CRT

Area of Science:

• Cardiology
• Medical Imaging
• Biomedical Engineering

Background:

• Cardiac resynchronization therapy (CRT) is a treatment for heart failure (HF) but has a significant non-responder rate.
• Accurate patient selection for CRT remains a challenge, impacting treatment efficacy.
• Echocardiography, specifically speckle tracking, shows promise for improving CRT candidate selection by assessing left ventricular (LV) mechanics.

Purpose of the Study:

• To determine the relative importance of strain-based echocardiographic features from different views (4, 3, and 2-chamber) for predicting CRT response.
• To identify which echocardiographic views provide the most informative features for CRT patient selection.
• To lay the groundwork for machine learning models to enhance CRT candidate selection.

Main Methods:

• Extraction of numerous strain-based features from longitudinal strain curves of 130 heart failure patients undergoing CRT.
• Application of feature selection techniques including out-of-bag random forest, wrapping, and filtering.
• Analysis of the importance of features derived from 4, 3, and 2-chamber echocardiographic views.

Main Results:

• Over 50% of the top 20 most important features for predicting CRT response were derived from the 4-chamber echocardiographic view.
• While less represented, strain features from the 2- and 3-chamber views provided valuable complementary information.
• Specific informative features were identified and analyzed for their contribution to predicting CRT outcomes.

Conclusions:

• The 4-chamber echocardiographic view is paramount for extracting key features that predict cardiac resynchronization therapy response.
• Integrating features from 2- and 3-chamber views can offer complementary insights, potentially improving predictive accuracy.
• This study represents a crucial step towards developing advanced machine learning tools for optimized CRT patient selection.