Predictive Value of Machine Learning-Based Echocardiographic Myocardial Texture After Transcatheter Aortic Valve...

Area of Science:

• Cardiology
• Medical Imaging
• Machine Learning

Background:

• Severe aortic valve stenosis necessitates treatment, with transcatheter aortic valve replacement (TAVR) offering a minimally invasive option.
• Predicting clinical endpoints after TAVR is crucial for patient management and optimizing outcomes.
• Echocardiographic myocardial texture analysis presents a novel avenue for risk stratification.

Purpose of the Study:

• To develop an interpretable machine learning (ML) model using echocardiographic myocardial texture.
• To predict 1-year clinical endpoints, specifically all-cause mortality and heart failure events, following transfemoral TAVR.
• To validate the model's performance using external data.

Main Methods:

• Retrospective analysis of 121 patients undergoing transfemoral TAVR, with external validation data.
• Application of nine ML algorithms to echocardiographic myocardial texture data.
• Development of an interpretable Extra Tree (ET) model utilizing Shapley additive explanations (SHAP) for feature importance.

Main Results:

• The Extra Tree (ET) model demonstrated superior discriminative ability among the nine ML algorithms.
• The final ET model accurately predicted 1-year clinical endpoints in both internal and external validation cohorts.
• Low left ventricular ejection fraction (LVEF <50%) was identified as an independent predictor of adverse outcomes; higher Rad-Score values correlated with reduced survival rates.

Conclusions:

• Echocardiographic myocardial texture analysis combined with ML offers a promising approach for predicting 1-year clinical endpoints post-TAVR.
• The developed interpretable ML model can serve as a valuable tool for clinical decision-making.
• This method facilitates enhanced risk stratification and personalized patient care following TAVR.