Time series to spatiotemporal HRV & hemodynamic assessment tool with blood pressure estimation

• Main Author: Chin, Ryan Ray Kang
• Other Authors: Ng Yin Kwee
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Engineering; Medicine, Health and Life Sciences
• Online Access: https://hdl.handle.net/10356/176145
• Institution: Nanyang Technological University

Cardiovascular diseases (CVD) are the leading cause of morbidity and mortality worldwide, influenced by genetics, environment, and lifestyle. Heart Rate Variability (HRV) serves as a critical tool for gauging autonomic regulation and cardiovascular health, offering insights into early cardiovascular stress and risk of CVD through non-invasive means. Unlike ECG which requires electrode placement, monitoring hemodynamics via Photoplethysmography (PPG) is also non-invasive but simpler, further aiding in early CVD detection. Key factors like monitoring blood pressure and blood sugar levels are essential in preventing CVD progression, which currently require inconvenient monitoring methods namely blood pressure cuffs and finger prick. This has spurred interest in developing more accessible, non-invasive technologies. This project will explore the above mentioned, to begin, Electrocardiogram (ECG) and Photoplethysmogram (PPG) signals are collect through a 3min data collection process from volunteers wearing a prototype wearable. The signals will go through a series of pre-processing before extracting the relevant data of the 4 main highlights of this project. (1) HRV analysis using HRV and CHA bandwidths which converts the time series of peak-to-peak interval (PPI) of PPG signals into a spatiotemporal form. Though ECG is conventionally used HRV analysis, PPG offers a non-invasive method of monitoring. Both bandwidth methods are analyzed with different groups namely age, gender, blood pressure, BMI, and the presence of medical conditions. The bandwidths also serve as a method for long term health and drug efficacy monitoring. (2) A blood pressure estimation was attempted using 11 features extracted from the ECG and PPG signals acquired from the data collection. Various Machine Learning (ML) and Neural Network (NN) models, including Physics-Informed Neural Networks (PINN), were attempted. The PINN model came up on top with an MAE of 4.80 and 3.66 mmHg for systolic and diastolic blood pressures respectively. The predictions also met the AAMI and BHS standards for BP measurement accuracy. (3) Plethysmorphometry (PMM), which is a 3D model of 2 or more PPG signals overlaid over one another, representing of the entire time series of PPG signals. The changes in hemodynamics or morphology of the PMM was analyzed on the Oral Glucose Tolerance Test (OGTT) volunteers. Findings of this project showed a less steep gradient following the systolic peak in the PMM and higher blood glucose levels (BGLs). (4) Finally, all components will be integrated and presented through a Graphic User Interface (GUI) application to simplify the overall analysis process.