A multi-domain feature-based ensemble learning approach for cuffless blood pressure estimation

Hypertension is one of the leading causes of premature death globally. A sudden surge in the blood pressure (BP) level creates an emergency and sometimes leads to severe consequences, including death. Generally, a sphygmomanometer is used to measure the BP. However, this technique has several limitations in continuous and long-term monitoring. Cuffless BP estimation has recently emerged as an excellent alternative to overcome these limitations, and can be used for real-time, remote, or ambulatory BP monitoring. Cuffless BP can be estimated using cardiovascular signals, including photoplethysmogram (PPG) and electrocardiogram (ECG) measurements. Several machine learning-based BP estimation methods have been proposed in the literature. Most of the techniques use a domain-specific signal feature to estimate the BP. This research proposes an efficient ensemble learning-based technique for estimating cuffless BP, utilizing multiple domain features, including temporal, spectral, and signal-specific features extracted from cardiovascular measurements. These multi-domain features are incorporated into the ensemble learning-based Extra Trees for robust BP estimation. Furthermore, the proposed method is successfully executed on the Xilinx PYNQ-Z2 platform to verify the hardware compatibility. The proposed method effectively estimates BP values with a mean absolute error (MAE) of 3.96 mmHg and 1.55 mmHg for SBP and DBP, respectively, outperforming several state-of-the-art methods. The proposed method satisfies various international standards related to the cuffless BP estimation task with good performance. • The research work presented in this paper involves the design, development, and evaluation of a new cuffless blood pressure measurement methodology. • This paper presents an effective ensemble learning-based technique that utilizes multi-domain features of ECG and PPG signals for cuffless BP estimation. The multidomain features employed in the Extra-Trees regressor (MDF-XT) algorithm are the temporal, spectral, and signal-specific information extracted from relevant cardiological measurements. • The proposed method satisfies several international standards, including Association for the Advancement of Medical Instrumentation (AAMI), British Hypertension Society (BHS), and Institute of Electrical and Electronics Engineers (IEEE) standards, with superior grading. • The proposed method is deployed on the Xilinx PYNQ-Z2 for a near-real-time cuffless BP estimation using an edge platform.