Real-time multi-class signal quality assessment of photoplethysmography using machine learning technique

Abstract

Photoplethysmography (PPG) signal quality assessment (SQA) ensures improved measurements of various surrogate cardiovascular measurements like heart rate, SpO2, blood pressure, cardiac output and many more and as well reduces false alrams in ambulatory measurements. Although PPG SQA (PSQA) is a well researched area, but multiclass prediction of signal quality and its hardware implementation is limited. In this paper, a new non-segmenting approach for multiclass PSQA is presented with an optimal set of seven time-frequency domain features in conjuction with supervised classifiers to identify clean, partly clean and corrupted PPG data. The present work was carried out in three major stages- signal (dataset) acquisition, feature extraction and 3-class classification using random forest classifier. The technique was evaluated with volunteers’ data, MIMIC-III data from PhysioNet, CSL data, and combined data and achieved an overall average accuracy of 96.8% with good (0.98 or above) sensitivity, specificity, F1 score, precision and recall. On-device implementation of the proposed PSQA was accomplished using 7.2 h of PPG dataset on a quad-core Broadcom BCM 2837 controller with 1.2 GHz, supported by 1 GB RAM-based standalone device, with an accuracy of 97.09%. The latency and the peak memory requirement of the implementation was 0.81 s and 19 kB respectively to process 3 s PPG data. The present technique may be integrated as a part of a remote healthcare system using wearable sensors.

A review: Blood pressure monitoring based on PPG and circadian rhythm

“…The model's performance was enhanced when trained on the processed data, leading to a reduction in the prediction error and an increase in the correlation between predicted and actual blood pressure values. Prasun et al 48 employed a combination of time and frequency domain analysis to derive seven distinct features, which were then integrated and applied with various ML classifiers for the purpose of classifying signals into three categories (clean, partially clean, and damaged). The final random forest (RF) classifier demonstrated extraordinary performance, achieving a mean accuracy of 96.8% in evaluations across four datasets.…”

Artifacts classification and apnea events detection in neck photoplethysmography signals

“…Following this approach, several machine learning algorithms have been proposed in the literature to discriminate artifacts from clean PPG. Examples of signal processing techniques used in these algorithms include: decision lists [39][40][41][42][43], decision trees [44,45], naïve Bayes classifiers [46], support vector machines (SVM) [36,[47][48][49][50], multi-layered perceptrons [51], personalized neural networks (NN) [52], and 1-D CNNs [53,54].…”

DNN based reliability evaluation for telemedicine data

“…Most previous studies presented results of detecting noise signals with very high accuracy (> 91%) using various deep learning techniques [ 8 – 12 ]. Roy et al, Lim et al, and Goh et al [ 8 – 11 ] have performed studies on the improvement of PPG signal quality based on waveform.…”

“…In addition, in deep learning-based signal quality determination, it might be difficult to improve signal quality because there is no accurate basis for judging abnormal data. Prasun et al [ 12 ] have proposed a method based on a feature extracted from PPG signal and determined the signal quality through seven feature sets including extracted kurtosis and entropy in time and frequency domains. In their study, it was possible to identify partially clean signals and noisy signals with a high accuracy (> 97%).…”