Integrated Noise-Aware PPG-Based Blood Pressure Stratification With Rate and Rhythm Sensing for Smart Portable Cuffless BP Monitor

AMDF + zero-crossing signal-quality assessment with pulse-rate estimation, onset/peak detection with rhythm recognition, then BP classification from 12 PPG features. 13 SQA approaches compared; 10 BP models tested on unseen databases (UCI, PulseDB, MIMIC, MIT-BIH SLP). Missed-alarm rates 7.4% (hypotension), 2.8% (normotension), 2.7% (hypertension); false-alarm reduction 7–100% on severe noise.

• Form factor: any

Blood pressure (BP), pulse rate (PR), and heart rhythm are all interconnected and need to be sensed simultaneously for the effective diagnosis and management of cardiovascular diseases, sepsis, stress, dehydration, and certain medications. Therefore, we propose an integrated noise-aware photoplethysmogram (PPG)-based BP stratification with PR and rhythm (RT) sensing, with the main objectives of reducing false alarms and improving the energy-efficiency of a cuffless ambulatory BP monitoring. The proposed method consists of a preprocessing, a highpass filter to remove baseline drifts, an average magnitude difference function (AMDF)-based signal quality assessment (SQA) with PR estimation, straightforward onset-peak detection with rhythm recognition, and BP classification.We presented 13 SQA approaches (four SQA approaches using AMDF and/or number of zero-crossing (NZC) features, eight machine learning (ML) SQA approaches using eight ML classifiers with horizontal visibility graph (HVG) feature), and a 1D-convolutional neural network (1D-CNN) model, and evaluated their performance using UCI, PulseDB, MIMIC, and MIT-BIH SLP databases to explore a lightweight SQA approach. We present ten BP models using the 12 PPG features and evaluate their performance using unknown databases by considering inter-subject and sensor variability. The noise-aware BP classification method had missed alarm rate (MAR) values of 7.4% (1753 HPT segments), 2.8% (89841 NT segments), 2.7% (181525 HT segments) and false alarm reduction rate (FARR) of 7%–100% (severe noisy PPG segments) under the motion-corrupted PPG signals, AWGN-corrupted PPG signals, and noise-free (NF) PPG signals taken from four standard databases. Using the best SQA approach [AMDF features and NZC feature from smoothed derivative PPG (dPPG)] and the best BP classifier [random forest (RF)], the proposed method had an overall BP classification accuracies of 77.0% [hypotension (HPT)], 85.4% [normotension (NT)], and 90.4% [hypertension (HT)], which is comparable to existing traditional ML and deep learning-based methods. The proposed integrated noise-aware PPG-based BP stratification with PR and RT sensing prototype using the Raspberry Pi computing platform can save energy from 6.3% (5-s noisy PPG segments) to 95.2% (60 s noisy PPG segments) when compared with an automated PPG analysis system without PPG quality checking. The code for our proposed system is available at https://github.com/sivanjaneyu/Noise-Aware-BPClassify

Verbatim from Semantic Scholar (opens in a new tab) · Abstract via Semantic Scholar; © the authors/publisher

No results from this paper on ChatPPG leaderboards yet. Browse the leaderboards.