Estimating Continuous Blood Pressure from Photoplethysmogram Signals for Non-invasive Devices by Convolutional Neural Network

Blood pressure (BP) is a crucial indicator of abnormal levels of stress on blood vessel walls, including hypertension or hypotension. Although several studies have attempted to predict BP based on photoplethysmogram (PPG), accuracy and resource consumption remain significant challenges. To address these challenges, we propose a flexible approach employing a deep convolutional neural network (CNN) with high accuracy. Our proposed deep-learning model delivers mean errors ± standard deviations of 0.18 ± 5.91 mmHg, -0,09 ± 3.21 mmHg, and 0.001 ± 3.82 mmHg for Systolic Blood Pressure (SBP), Diastolic Blood Pressure (DBP), and Mean Arterial Pressure (MAP), respectively. Furthermore, our method meets the standards set by the Advancement of Medical Instrumentation (AAMI) and achieves an "A" grade performance, as required by the British Hypertension Society (BHS) standard. Compared with previous benchmarks, our model achieves greater accuracy with fewer parameters, offering potential for real-time non-invasive BP monitoring using wearable devices.