A multicenter study shows machine learning algorithms improve heart failure detection using phonocardiogram and ECG data, suggesting clinical utility for early intervention.
Key Points
This study aims to validate an AI-based algorithm for early detection of heart failure using ECG and phonocardiogram data.
Multicenter study with 1,017 individuals divided into training and validating cohorts.
Used dimensionality reduction with the least absolute shrinkage and selection operator for model construction.
Evaluated multiple machine learning algorithms including logistic regression and random forest for screening heart failure.
302 participants reported heart failure out of the total.
The CatBoost model achieved an area under the curve of 0.998, indicating high accuracy.
Sensitivity and specificity of the CatBoost model were both 0.989, demonstrating strong detection performance.