Abstract:Comprehending the regulatory mechanisms influencing blood pressure control is pivotal for continuous monitoring of this parameter. Implementing a personalized machine learning model, utilizing data-driven features, presents an opportunity to facilitate tracking blood pressure fluctuations in various conditions. In this work, data-driven photoplethysmograph features extracted from the brachial and digital arteries of 28 healthy subjects were used to feed a random forest classifier in an attempt to develop a system capable of tracking blood pressure. We evaluated the behavior of this latter classifier according to the different sizes of the training set and degrees of personalization used. Aggregated accuracy, precision, recall, and F1-score were equal to 95.1%, 95.2%, 95%, and 95.4% when 30% of a target subject's pulse waveforms were combined with five randomly selected source subjects available in the dataset. Experimental findings illustrated that incorporating a pre-training stage with data from different subjects made it viable to discern morphological distinctions in beat-to-beat pulse waveforms under conditions of cognitive or physical workload.

What problem does this paper attempt to address?

The problem this paper attempts to address is how to detect blood pressure changes related to physical or cognitive load through a personalized multi-classification model. Specifically, the authors developed a random forest classifier based on photoplethysmography (PPG) signals to monitor blood pressure fluctuations in healthy subjects in real-time under different conditions (such as rest, cognitive tasks, and post-exercise). The main goal of the paper is to improve the accuracy of detecting individual blood pressure changes through a data-driven approach, thereby achieving non-invasive, continuous blood pressure monitoring. ### Background of the Paper With the growth of the population and the aging trend, the incidence of cardiovascular diseases (CVDs) is continuously rising. Hypertension (i.e., elevated blood pressure) is a medical condition that promotes coronary artery disease and other pathological conditions affecting vital organs such as the brain and kidneys. To overcome the limitations of traditional invasive blood pressure measurement methods, non-invasive blood pressure measurement systems have emerged. These devices use various sensors to extract information about the cardiovascular system's state through different physical principles. Continuous arterial blood pressure measurement is considered a more accurate indicator for assessing cardiovascular risk because changes in the blood pressure waveform reflect alterations in arteries and arterioles. ### Research Methods 1. **Data Collection**: - A custom-designed data acquisition system was used to obtain raw PPG pulse waveforms from the brachial and digital arteries of 28 healthy subjects in real-time. - The experiment was divided into three main parts: the rest stage (REST), the cognitive test stage (CT), and the post-exercise stage (AE). Data was collected three times for one minute in each stage, and a commercial cuff-based device, BPM Connect, was used as a reference standard for blood pressure measurement. 2. **Data Processing**: - Data preprocessing included band-pass filtering (0.3 Hz to 15 Hz) to remove DC offset and high-frequency noise. - Signals potentially contaminated by motion artifacts were identified and removed by segmenting into 3-second non-overlapping windows. - Pulse wave quality assessment (PWQA) was performed using a template matching method, and pulse waves that did not meet quality standards were marked and discarded. - Features were extracted by identifying key reference points (such as systolic peaks, dicrotic notches, and diastolic peaks). 3. **Model Training**: - The performance of 10 different personalized models (PSMs) and a person-independent model (PIM) was compared on the same dataset. - A random forest classifier was used, and model performance was optimized through a Leave-One-Subject-Out cross-validation strategy. - The amount of training data for personalized models and the proportion of target subject data were gradually increased to evaluate the effects of different levels of personalization. ### Main Contributions 1. Proposed 10 pre-training strategies to calibrate the random forest model based on individual physiological characteristics. 2. Pre-training significantly improved the accuracy of blood pressure classification for each pulse waveform, increasing it by 60% compared to the general model. 3. By increasing the number of source subjects, the risk of overfitting was reduced, and the need for additional target subject data was minimized. 4. The amount of data required for personalized models was reduced by 30% while maintaining evaluation metrics above 95%. ### Results - After data processing, a total of 17,886 pulse waveforms were obtained for analysis. - Personalized models performed excellently in terms of accuracy, precision, recall, and F1 score. The model's performance further improved, especially after the pre-training stage that combined data from different subjects. ### Conclusion This study demonstrates that personalized multi-classification models can effectively detect blood pressure changes related to physical or cognitive load, providing a new technical means for non-invasive, continuous blood pressure monitoring.

Development of a Personalized Multiclass Classification Model to Detect Blood Pressure Variations Associated With Physical or Cognitive Workload

Non-Invasive Risk Stratification of Hypertension: A Systematic Comparison of Machine Learning Algorithms

Synchronized wearables for the detection of haemodynamic states via electrocardiography and multispectral photoplethysmography

Continuous PPG-Based Blood Pressure Monitoring Using Multi-Linear Regression

Blood pressure stratification using photoplethysmography and light gradient boosting machine

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

Developing Personalized Models of Blood Pressure Estimation from Wearable Sensors Data Using Minimally-trained Domain Adversarial Neural Networks

Blood Pressure Estimation Using Photoplethysmography Only: Comparison between Different Machine Learning Approaches

A Multi-scenario Attention-based Generative Model for Personalized Blood Pressure Time Series Forecasting

Personalized Modeling of Blood Pressure with Photoplethysmography: an Error-Feedback Incremental Support Vector Regression Model

Using Photoplethysmography to Detect Real-time Blood Pressure Changes with a Calibration-free Deep Learning Model

A Refined Blood Pressure Estimation Model Based on Single Channel Photoplethysmography.

Accurate and 30-plus days reliable cuffless blood pressure measurements with 9-minutes personal photoplethysmograph data and mixed deduction learning

Detection and Classification of Unannounced Physical Activities and Acute Psychological Stress Events for Interventions in Diabetes Treatment

Systolic Blood Pressure Estimation from PPG Signal Using ANN

Quality Assessment of Photoplethysmography Signals For Cardiovascular Biomarkers Monitoring Using Wearable Devices

Photoplethysmography and Deep Learning: Enhancing Hypertension Risk Stratification

A continuous cuffless blood pressure measurement from optimal PPG characteristic features using machine learning algorithms

An IoMT-Driven Framework for Precision Cardiovascular Assessment Incorporating Multiscale Perspectives and Microfiber Bragg Grating

Personalized Stress Monitoring using Wearable Sensors in Everyday Settings

Study of cuffless blood pressure estimation method based on multiple physiological parameters