Yen, Liang-WeiLiang-WeiYenLuo, Zheng-YuZheng-YuLuoHuang, Jiun-WoeiJiun-WoeiHuangLee, Shu-ShengShu-ShengLeeHsiang-Chieh LeeCHIH-KUNG LEE2025-09-222025-09-222025https://www.scopus.com/record/display.uri?eid=2-s2.0-105014762747&origin=resultslisthttps://scholars.lib.ntu.edu.tw/handle/123456789/732248In this study, we propose an automatic optical system for continuous blood pressure (BP) monitoring which can calculate pulse transit time (PTT) using photoplethysmography (PPG) and image-photoplethysmography (iPPG) signals, and incorporating key physiological parameters such as blood volume, changes in arterial diameter, and heart rate variability (HRV) to enhance BP measurement accuracy. Given the widespread use of smartphones and headphones, we selected the ear as the measurement site for PPG while using a camera to capture the iPPG signals for the wrist radial artery. Additionally, since the radial artery is at the front of the wrist and images may include other areas beyond this region, we developed an algorithm for automatically identifying the wrist’s region of interest (ROI), replacing the need for manual operation. Then, we utilized a Random Forest model to classify the iPPG signals into three categories and to compute a weighted average to represent the overall iPPG signal of the wrist radial artery. In order to extract waveform features accurately, the ear PPG signal was used as a reference to build a CNN-BiLSTM + Global Context block (GC block) model that enhances the stability of the iPPG waveform. In addition, a features extraction algorithm was implemented for the PPG/iPPG signals, particularly those with less prominent characteristics. Our results demonstrated that this system has the potential to significantly improve personal healthcare, surpassing and as an improvement to traditional cuff-based blood pressure monitors.continuous blood pressure monitoringheart rate variabilityimage processingmachine learningpulse transit time[SDGs]SDG3conference paper10.1117/12.3050329