Wearable photoplethysmographic signals for cardiovascular disease diagnostics

• Main Author: Ng, Zhen Yuan
• Other Authors: Ng Yin Kwee
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Bioengineering; Engineering::Mechanical engineering
• Online Access: https://hdl.handle.net/10356/150703
• Institution: Nanyang Technological University

This report presents the development and application of a prototype for non-invasive, cuffless blood pressure (BP) monitoring. Two initial prototypes were designed – one to capture a single photoplethysmographic (PPG) signal for waveform analysis, and the other to capture dual PPG signals for pulse transit time (PTT) analysis. For the initial design of the dual PPG prototype, there was a problem with the time lag of the 2 PPG signals which could potentially deviate the PTT by a few milliseconds. To overcome this problem, a tentative design using a new Analog Front End AFE4900 was selected, with an analog-to-digital converter (ADC) that can process synchronised signals up to four PPG + one electrocardiogram (ECG). Using the existing single PPG workable prototype, its applications were tested by assessing its performance for measuring heart rate, oxygen saturation (SpO2) and BP estimation. The test showed that the prototype can obtain a satisfactory measurement for heart rate and SpO2 after comparing with a clinical-grade device. The BP estimation was carried out by first obtaining PPG and BP readings from 15 volunteers using the prototype. Four morphological PPG waveform features were extracted for training and a decision tree-based ensemble learning – Random Forest model was used. The estimated BP showed a very poor model performance with high mean squared error across the cross-validation (10-fold), training and test sets. Lastly, to improve PPG signals for a wearable device, motion artifact (MA) reduction method was studied using the frequency of motion from the accelerometer inputs. This frequency was used as the rejection frequency of a band stop filter in an attempt to remove the MA distortions in the signal within a fixed time window. This method was able to improve the peak detection of PPG signals when simple motion was introduced. When the random motion was introduced, a reduction in the time window is required to improve the peak detection. This project demonstrated that the current prototype could serve as a feasible design for PPG acquisition and MA reduction. However, the results obtained from these several experiments suggest that further studies such as increasing datasets and model’s complexity for BP estimation and other methods to deal with MA are still needed to be explored on to create a robust wearable device for cuffless, non-invasive BP monitoring.