Research On The Technology Of Non-invasive Hemoglobin Measurement Based On Photoplethysmography

Non-invasive hemoglobin concentration detection plays an important role in operation, blood transfusion,blood donation and nutrition examination, etc. The traditional method for measuring hemoglobin concentration needs to draw blood from human. It not only aggravates the physiological and psychological stress of the subject but also cannot monitor the variation of hemoglobin concentration continuously and real-timely. Until now, it is still a challenge to find a proper method and implementation for monitoring hemoglobin concentration continuously, non-invasively and real-timely. This paper focuses on researching the technique of continuous, non-invasive and real-time monitoring hemoglobin concentration and its implementation. The main work and achievements of this paper are as follows.We introduced the basic optical and physiological knowledge about the technology of non-invasive hemoglobin measurement. In according to the Lambert-Beer law we proposed the mathematical model to non-invasively measure hemoglobin concentration based on multichannel PPG signals.We constructed multichannel PPG signals capture system, including the probe, the front-end signal capturing circuit and the upper computer software. By using this developed multichannel PPG signal capture system, 324 samples were successfully captured, providing the reliable data for the subsequent PPG signals processing and hemoglobin concentration prediction.We studied on the PPG signal processing methods, including filtering out the high-frequency noise, reducing the motion artifact interference, removing the baseline drift from PPG signals and extracting the feature information from PPG signals. To filter out the high-frequency nosie, we applied the FIR filter, IIR filter and moving average filter. By processing the PPG signal using these three methods, all of them could effectively filter out the high-frequency noise. To reduce the motion artifact from PPG signal, we developed two new methods: one is combining the constrained independent component analysis and adaptive filter, the other is the method based on comb filter. The experiments on the synthetic dataset and real-world PPG signals demonstrate that both of the methods could reduce the motion artifact effectively. To remove the baseline drift, we studied three methods: median filtering, wavelet transform and Hilbert-Huang transform. All these three methods could remove the baseline drift from PPG signals to some extent.We captured the experimental data and implemented the regression modeling between the feature information extracted from PPG signals and hemoglobin concentration obtained by the complete blood cell counter. By using the PLS regression algorithm(n=153), the root mean square error(RMSE) is 1.56 g/dL and the correlation coefficient is 0.56(p<0.01). By using the PCA combining BP artificial neural network(ANN)(n=62), the RMSE is 1.49 g/dL and the correlation coefficient is 0.61(p<0.01). The results suggest that the regression model could predict the hemoglobin concentration well. It also demonstrates that it could non-invasively measure the hemoglobin concentration by using the developed system based on multiple channel PPG signals.