Estimating Blood Pressure from Photoplethysmogram Signal and Demographic Features using Machine Learning Techniques

Abstract: Hypertension is a potentially unsafe health ailment, which can be indicated directly from the Blood pressure (BP). Hypertension always leads to other health complications. Continuous monitoring of BP is very important; however, cuff-based BP measurements are discrete and uncomfortable to the user. To address this need, a cuff-less, continuous and a non-invasive BP measurement system is proposed using Photoplethysmogram (PPG) signal and demographic features using ML algorithms. PPG signals were acquired from 219 subjects, which undergo pre-processing and feature extraction steps. Time, frequency and time-frequency domain features were extracted from the PPG and their derivative signals. Feature selection techniques were used to reduce the computational complexity and to decrease the chance of over-fitting the ML algorithms. The features were then used to train and evaluate ML algorithms. The best regression models were selected for Systolic BP (SBP) and Diastolic BP (DBP) estimation individually. Gaussian Process Regression (GPR) along with ReliefF feature selection algorithm outperforms other algorithms in estimating SBP and DBP with a root-mean-square error (RMSE) of 6.74 and 3.59 respectively. This ML model can be implemented in hardware systems to continuously monitor BP and avoid any critical health conditions due to sudden changes.

• The paper proposes a cuff-less method for blood pressure estimation by combining Photoplethysmogram (PPG) signals with demographic features using machine learning.
• Utilizing extensive feature sets and selection methods, Gaussian Process Regression (GPR) and Ensemble Trees models achieved the lowest RMSE (6.74 for SBP, 3.59 for DBP) when incorporating demographic data.
• This research has significant implications for continuous blood pressure monitoring in wearable technology and lays groundwork for future non-invasive diagnostic advancements.

Estimating Blood Pressure from Photoplethysmogram Signal and Demographic Features using Machine Learning Techniques

The study presented in this paper explores the integration of Photoplethysmogram (PPG) signals and demographic data for the estimation of blood pressure (BP) using ML algorithms. The primary motivation behind this research is the development of a cuff-less, non-invasive, and continuous method for BP monitoring, addressing the limitations and discomfort associated with traditional cuff-based measures.

The dataset utilized comprises PPG signals from 219 subjects and undergoes rigorous preprocessing steps including feature extraction across time-domain, frequency-domain, and statistical dimensions. The study introduces a comprehensive suite of features—75 time-domain, 16 frequency-domain, and 10 statistical features—designed to leverage the intricate dynamics of PPG signals. Additionally, demographic variables such as age, height, weight, BMI, and heart rate are included, enriching the feature set.

For feature selection, the study employs methods like correlation-based feature selection (CFS), ReliefF, and the minimum redundancy maximum relevance (fscmrmr) algorithm, aimed at reducing overfitting while maintaining model robustness. Gaussian Process Regression (GPR) and Ensemble Trees emerged as the superior models, with GPR, when combined with ReliefF feature selection, yielding the lowest root mean square error (RMSE) of 6.74 for systolic blood pressure (SBP) and 3.59 for diastolic blood pressure (DBP).

A distinctive element of this paper is its comparative analysis of different ML models on both existing and novel feature sets. The incorporation of demographic features appears to substantially enhance BP estimation accuracy, as demonstrated by the strong correlation coefficients of 0.95 for SBP and 0.96 for DBP upon hyperparameter optimization.

The implications of this research are substantial, particularly for wearable health technology. The incorporation of such algorithms into portable devices paves the way for continuous real-time BP monitoring, potentially mitigating risks associated with hypertensive events through early detection. The approach also lays foundational knowledge for advancing non-invasive diagnostic methods, thereby fostering further research in ML-based healthcare applications.

Looking forward, the refinement of this methodology with larger datasets, incorporating deep learning techniques, could improve the model's predictive performance towards fulfilling regulatory standards such as the International Organization for Standardization (ISO) and the British Hypertension Society (BHS) criteria. This advancement holds promise for not only raising the bar for accuracy in BP estimation but also making such technologies more accessible in everyday healthcare solutions.