METHODS: Continuous raw PPG waveforms were blindly allocated into segments with an equal length (5s) without leveraging any pulse location information and were normalized with Z-score normalization methods. A 1-D-CNN was designed to automatically learn the intrinsic features of the PPG waveform, and perform the required classification. Several training hyperparameters (initial learning rate and gradient threshold) were varied to investigate the effect of these parameters on the performance of the network. Subsequently, this proposed network was trained and validated with 30 subjects, and then tested with eight subjects, with our local dataset. Moreover, two independent datasets downloaded from the PhysioNet MIMIC II database were used to evaluate the robustness of the proposed network.
RESULTS: A 13 layer 1-D-CNN model was designed. Within our local study dataset evaluation, the proposed network achieved a testing accuracy of 94.9%. The classification accuracy of two independent datasets also achieved satisfactory accuracy of 93.8% and 86.7% respectively. Our model achieved a comparable performance with most reported works, with the potential to show good generalization as the proposed network was evaluated with multiple cohorts (overall accuracy of 94.5%).
CONCLUSION: This paper demonstrated the feasibility and effectiveness of applying blind signal processing and deep learning techniques to PPG motion artifact detection, whereby manual feature thresholding was avoided and yet a high generalization ability was achieved.
METHODS: Fifty-one participants performed the standard incremental treadmill exercise in a controlled laboratory setting with 12-lead ECG attached to the patient's body and wearing wrist-worn PPG trackers.
RESULTS: At each stage, the absolute percentage error of the PPG was within 10% of the standard acceptable range. Further analysis using a linear mixed model, which accounts for individual variations, revealed that PPG yielded the best performance at the baseline low-intensity exercise. As the stages progressed, heart rate validity decreased but was regained during recovery. The reliability was moderate to excellent.
CONCLUSIONS: Low-cost trackers AMAZFIT Cor and Bip validity and reliability were within acceptable ranges, especially during low-intensity exercise among patients with ischemic heart disease recovering from cardiac procedures. Though using the tracker as part of the diagnosis tool still requires more supporting studies, it can potentially be used as a self-monitoring tool with precautions.
APPROACH: Several master templates are initially generated by applying principal component analysis to data obtained from the PhysioNet MIMIC II database. The master template is then updated with each incoming clean PPG pulse. The correlation coefficient is used to classify the PPG pulse into either good or bad quality categories. The performance of our algorithm was evaluated using data obtained from two different sources: (i) our own data collected from 19 healthy subjects using the wearable Sotera Visi Mobile system (Sotera Wireless Inc.) as they performed various movement types; and (ii) ICU data provided by the PhysioNet MIMIC II database. The developed algorithm was evaluated against a manually annotated 'gold standard' (GS).
MAIN RESULTS: Our algorithm achieved an overall accuracy of 91.5% ± 2.9%, with a sensitivity of 94.1% ± 2.7% and a specificity of 89.7% ± 5.1%, when tested on our own data. When applying the algorithm to data from the PhysioNet MIMIC II database, it achieved an accuracy of 98.0%, with a sensitivity and specificity of 99.0% and 96.1%, respectively.
SIGNIFICANCE: The proposed method is simple and robust against individual variations in the PPG characteristics, thus making it suitable for a diverse range of datasets. Integration of the proposed artefact detection technique into remote monitoring devices could enhance reliability of the PPG-derived physiological parameters.