Abstract
Objective: Accurate and continuous monitoring of arterial blood pressure (BP) is essential for effective management of cardiac and hypertension-related conditions. The limitations of non-invasive measurement techniques and the insufficient accuracy of invasive methods highlight the need for novel approaches. This study introduces a machine learning-based methodology for precise and continuous BP estimation by analyzing electrocardiography (ECG) signals in sub-frequency bands. Methodology: ECG and BP recordings from five subjects, sourced from the IEEE database, were utilized. The signals were segmented into specific sub-frequency bands (e.g., 2-40 Hz, 5-15 Hz), resulting in 594 machine learning models. A 10-second epoching process was applied to the ECG signals to enhance prediction accuracy. Statistical features were extracted from the time domain, and correlation analysis was performed to identify BP-relevant features for model training. Gaussian process regression, decision tree, and ensemble bagged tree algorithms were employed to develop and evaluate the prediction models. Results: Among the tested models, the best performance was achieved using the 2-40 Hz frequency band of the 10-second epoch ECG signals, yielding mean absolute percentage errors (MAPE) of 1.80 for systolic BP and 1.36 for diastolic BP, with a coefficient of determination (R2) of 0.99. Conclusion: This study demonstrates that analyzing ECG signals within specific frequency ranges strengthens their correlation with arterial blood pressure. Moreover, the inclusion of epoched ECG signals significantly enhances the predictive accuracy of BP estimation models, providing a promising foundation for non-invasive BP monitoring applications.
Funding Information
  • Sakarya University Scientific Research Projects Coordination Unit within the scope of the Graduate Thesis Projects (2022-7-24-125)

This publication has 52 references indexed in Scilit: