Fig 1.
An illustration of the overall procedure followed in this study.
Table 1.
Clinical characteristics of the heart failure patients based on their LVEF categories.
Table 2.
Definitions of ECG features.
Table 3.
Regression models and hyperparameters tuning.
Table 4.
ECG features for the three groups (HFpEF, HFmEF, and HFrEF).
Table 5.
Average RMSE values through 24-Hours, using regression models with best hyperparameters.
Fig 2.
RMSE per hour using the best hyperparameters combination for the four regression models.
Red circles show the hours of occurrence of the lowest RMSE values. Where QTc and QRS are found to be the most important features for evaluating the LVEF levels.
Fig 3.
Original and estimated LVEF values per patient for hours (22:00–23:00, 08:00–09:00, 00:00–01:00) using the Tree, GPR, XGBOOST and SVM, respectively.
Fig 4.
Correlation plots between the original and estimated LVEF values for hours (22:00–23:00, 08:00–09:00, 00:00–01:00) using the Tree, GPR, XGBOOST and SVM, respectively.
Fig 5.
Bland-Altman plots between the average value of (Original and estimated LVEF) and their corresponding difference for hours (22:00–23:00, 08:00–09:00, 00:00–01:00) using the Tree, GPR, XGBOOST and SVM, respectively.
Fig 6.
Feature importance over the Tree regression model (10:00–11:00pm).
This study used normalized importance scores for the ECG features (QTc, QRS, ST-T, TP, Entropy, and Instant Frequency) to estimate the LVEF levels in the three HF categories. Indicating that QTc is the most important feature for evaluating the LVEF levels.
Fig 7.
Correlation matrix of the RMSE values through 24-Hours, using the best regression model.
Black asterisks show the hours of statistical significance to circadian changes.
Table 6.
Summary of the studies for LVEF estimation in preserved, midrange, and reduced HF patients.
Best performance highlighted in yellow.