Camera-Based Neonatal Blood Pressure Estimation From Multisite and Multiwavelength Pulse Transit Time—A Proof of Concept in NICU

Blood pressure (BP) is a vital physiological parameter for early warning and prompt intervention treatment in the neonatal intensive care unit (NICU). However, the application of contactless BP measurement methods in neonates remains under-explored. This proof-of-concept clinical study proposes using multi-site and multi-wavelength pulse transit time (PTT) generated from remote-Photoplethysmography (rPPG) for neonatal BP estimation. A dataset of 40 neonates was created in the NICU under three alternating phases (resting -BP measurement -resting). The spatially averaged rPPG signals from five body parts were used to calculate multiple PTT features, including multi-site PTT (MS-PTT) derived from different body parts and multi-wavelength PTT (MW-PTT) derived from different skin layers, for BP estimation. Three machine learning models, including Multivariate Linear Regression (MLR), Support Vector Regression (SVR), and Random Forest Regression (RFR), were employed for both univariate and multivariate regression. Combining MS-PTT and MW-PTT yielded the best results, achieving a mean absolute error ± standard deviation (MAE ± STD) of 7.65 ± 7.48 mmHg for SBP, 6.31 ± 5.58 mmHg for DBP, and 7.29 ± 7.29 mmHg for MBP, based on MLR with subject-dependent modeling. According to the British Hypertension Society guidelines, these results meet the requirements for Grade C. These findings provide the first clinical proof-of-concept of using camera-based MS-PTT and MW-PTT features for contactless neonatal BP estimation.