Computational method from embedded wearable PPG for BP/BGL prediction

• Main Author: Muk, Nathanael Chen Han
• Other Authors: Ng Yin Kwee
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Engineering; Mathematical Sciences; Medicine, Health and Life Sciences; Machine learning; Blood pressure
• Online Access: https://hdl.handle.net/10356/177491
• Institution: Nanyang Technological University

As cardio-metabolic diseases grow in prevalence, current management and monitoring methods fail to address some of the many inconveniences and difficulties that have existed for a long time. The traditional approaches to blood pressure and blood glucose monitoring are adequately sufficient in producing precise results that can be relied on clinically. Yet, these traditional methods, such as the sphygmomanometer and the glucose prick test, are invasive and non-ambulatory in nature. As we seek to improve the lives of patients who already face the difficult task of managing these diseases, more can be done to bridge the gap between such clinically accurate devices and more convenient, less reliable devices such as smartwatches. Many studies have explored the different ways in which alternative methods of blood pressure and glucose measurements can be obtained. In this report, photoplethysmography will be the main point of focus. Its use as a means of providing features useful in determining blood pressure and blood glucose levels will be explored by highlighting its past successes and limitations. Various ways in which photoplethysmography signals can be processed and analysed are discussed featuring a sub-study into pharmacological effects on photoplethysmogram morphology. Additionally, machine learning techniques will be studied as a means of improving the reliability and accuracy of these processing methods (signal filtering, normalizing, feature detection, etc), in the hopes of paving a way forward in addressing some of the common limitations in the photoplethysmography approach. By addressing the limitations common to wearables, such as noise, motion artifacts, loss of features, this study seeks to improve on the processing and quality of photoplethysmography signals. Ultimately, improved processing and quality opens opportunities for more robust and reliable work to be done, especially in the area of utilising photoplethysmography signals as inputs for predictive machine learning tasks in blood pressure and blood glucose level estimation.