China-calibration-ready echocardiographic feature modeling for LVEF-derived dysfunction classification: a...

Area of Science:

• Cardiology
• Medical Imaging
• Machine Learning

Background:

• Anthracycline chemotherapy can cause cardiotoxicity, impacting right ventricular function.
• Accurate prediction of cardiotoxicity is crucial for patient management.
• Echocardiography is a key imaging modality for assessing cardiac function.

Purpose of the Study:

• To present a reproducible echocardiographic feature-engineering and machine-learning pipeline.
• To classify a surrogate label for left ventricular dysfunction derived from left ventricular ejection fraction (LVEF).
• To serve as a methodological proof-of-concept for future clinical applications.

Main Methods:

• Utilized a dataset of 1,000 two-dimensional echocardiographic tracings from public databases.
• Engineered twelve geometric and morphometric features, including left ventricular trace-area and bounding-box area changes.
• Trained five classifiers, including logistic regression, using stratified splits and cross-validation.

Main Results:

• Logistic regression achieved a high ROC-AUC of 0.996 and F1 score of 0.966.
• Performance metrics reflect feature-label coupling, not independent clinical prediction.
• Left ventricular trace-area change mediated the association between bounding-box area change and dysfunction.

Conclusions:

• The study provides a transparent, reusable analytical scaffold for future research.
• The pipeline is a proof-of-concept, not a clinically validated predictive model.
• Future work should incorporate prospective data with true right ventricular outcomes for cardiotoxicity prediction.