Estimating blood pressure trends and the nocturnal dip from photoplethysmography
| Autor: | Mustafa Radha, Ronald M. Aarts, Nikita B. Rajani, Nathalie Velthoven, Nadja Kobold, Valentina Vos, Petra A. Wark, Pedro Fonseca, Nikolaos Mastellos, Koen de Groot, Cybele Cp Wong, Reinder Haakma |
|---|---|
| Přispěvatelé: | Signal Processing Systems, Biomedical Diagnostics Lab, Center for Care & Cure Technology Eindhoven |
| Jazyk: | angličtina |
| Rok vydání: | 2019 |
| Předmět: |
Male
Technology Physiology PULSE TRANSIT-TIME DEVICE 02 engineering and technology physics.med-ph NORMALIZATION Engineering 0302 clinical medicine 0903 Biomedical Engineering NIGHT Heart rate variability Mathematics RISK Artificial neural network Signal Processing Computer-Assisted Middle Aged neural networks Circadian Rhythm Random forest VARIABILITY 0906 Electrical and Electronic Engineering Female Life Sciences & Biomedicine Adult Adolescent Mean squared error Systole 0206 medical engineering Biomedical Engineering Biophysics FOS: Physical sciences HEART-RATE Young Adult 03 medical and health sciences Deep Learning Physiology (medical) Photoplethysmogram Linear regression Humans Engineering Biomedical ambulatory blood pressure Models Statistical Science & Technology business.industry Blood Pressure Determination Pattern recognition SLEEP Physics - Medical Physics 020601 biomedical engineering Data set Blood pressure 1116 Medical Physiology free-living protocol photoplethysmography Medical Physics (physics.med-ph) Artificial intelligence business 030217 neurology & neurosurgery |
| Zdroj: | Physiological Measurement, 40(2):025006. Institute of Physics |
| ISSN: | 0967-3334 |
| Popis: | Objective: Evaluate a method for the estimation of the nocturnal systolic blood pressure dip from 24-hour blood pressure trends using a wrist-worn Photoplethysmography (PPG) sensor and a deep neural network in free-living individuals, comparing the deep neural network to traditional machine learning and non-machine learning baselines. Approach: A wrist-worn PPG sensor was worn by 106 healthy individuals for 226 days during which 5111 reference values for blood pressure were obtained with a 24-hour ambulatory blood pressure monitor as ground truth and matched with the PPG sensor data. Features based on heart rate variability and pulse morphology were extracted from the PPG waveforms. Machine learning models (linear regression, random forests, dense neural networks and long- and short-term memory neural networks) were then trained and evaluated in their capability of tracking trends in systolic and diastolic blood pressure, as well as the estimation of the nocturnal systolic blood pressure dip. Main results Best performance was obtained with a deep long- and shortterm memory neural network with a Root Mean Squared Error (RMSE) of 3.12±2.20 ∆mmHg and a correlation of 0.69 (p = 3 ∗ 10−5) with the ground truth Systolic Blood Pressure (SBP) dip. This dip was derived from trend estimates of blood pressure which had an RMSE of 8.22±1.49 mmHg for systolic and 6.55±1.39 mmHg for diastolic blood pressure. The random forest model showed slightly lower average error magnitude for SBP trends (7.86±1.57 mmHg), however Bland-Altmann analysis revealed systematic problems in its predictions that were less present in the long- and short-term memory model. Significance The work provides first evidence for the unobtrusive estimation of the nocturnal blood pressure dip, a highly prognostic clinical parameter. It is also the first to evaluate unobtrusive blood pressure measurement in a large data set of unconstrained 24-hour measurements in free-living individuals and provides evidence for the utility of long- and short-term models in this domain. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|