|
KMID : 1137820210420020031
|
|
의공학회지
2021 Volume.42 No. 2 p.31 ~ p.38
|
|
|
Reliability of the Photoplethysmographic Analysis Using Deep Neural Network (DNN) Algorithm
|
|
Kim Ji-Woon
Park Sung-Min
Choi Seong-Wook
|
|
|
Abstract
|
|
|
|
In this study, the waveform parameters of photoplethysmography were obtained by using a new algorithm incorporating deep neural networks (DNN) and compared data measured from electrocardiogram and parameters manually acquired by experts, to analyze the accuracy and errors of the suggested algorithm. The 6 DNNs of the algorithm had been trained to find the onset, systolic, W and Z points and to determine the start and end of noninformative regions. The algorithm inputs 3 seconds of PPG data measured in real time into DNNs, records the position of the DNNs’ output. The final decision accuracy of the algorithm was improved by using repeating number of DNNs’ output when overlapping the input data section and the output section. HR and HRV obtained by the algorithm showed 87.7% and 86.7% agreement compared to those determined by ECG. Aix, crest time, Pulse Propagation Time (PPT) among PPG parameters showed 0.080±0.15 mV, -0.012±0.030 sec and 0.033±0.109 sec error compared to those manually measured by human experts. The algorithm proposed in this study was able to quickly provide accurate PPG parameters with little errors even when noise and arrhythmia occurred.
|
|
|
KEYWORD
|
|
|
Photoplethysmography (PPG), Pulse waveform analysis (PWA), Deep neural network (DNN), Recognition Score (RS)
|
|
|
FullTexts / Linksout information
|
|
|
|
|
|
Listed journal information
|
|
 
|