Abstract
Cardiac resynchronisation therapy (CRT) is a successful treatment for patients with chronic heart failure. However, 30–50 % of patients do not respond to the therapy, and novel predictive tools are needed to improve patient selection. In this pilot study, we proposed novel features of vectorcardiogram (VCG) signals derived from the principal component analysis (PCA) and evaluated their ability to predict response to CRT. Retrospective data from 47 patients who underwent CRT device implantation were used. For each patient, the 12-lead ECG was recorded and the VCG was reconstructed using the Kors' regression. In addition to conventional indices characterising the area of QRS complex, we performed PCA of VCG signals to characterise the VCG loop shape and complexity. Linear supervised machine learning models were fed with VCG features to predict CRT response defined as more than 5% increase in left ventricular ejection fraction in a year after implantation. The best performing Support Vector Machine model (ROC AUC 0.75, F1-score 0.75, precision 0.78, recall 0.72) utilized two novel VCG features: Shape Planarity index as a measure of the VCG loop deviation from the main PCA plane and PCA1 index characterising the VCG loop elongation along the first PCA axis. The Shape Planarity index showed the highest significance in predicting CRT response among the tested features. Our results suggest a great potential of PCA based VCG features in predicting CRT response.
Funding Information
  • Russian Science Foundation (19-14-00134)

This publication has 16 references indexed in Scilit: