Finger photoplethysmogram signal enhancement: Comparing performance between PCA and ICA methods
01 Apr 2017-pp 203-209
TL;DR: This paper compares the performance between two popular statistical signal processing tools, viz., principal component analysis (PCA) with fast independent component analysis (/ICA) in reduction of MA from finger pulse signal collected from 30 human volunteers and found that beat to beat correlation is higher in the PCA preprocessed data.
Abstract: Pulse signal is prone to corruption with motion artifacts (MA) due to attachment of the sensor to extreme body parts like finger, toes and forehead. This paper compares the performance between two popular statistical signal processing tools, viz., principal component analysis (PCA) with fast independent component analysis (/ICA) in reduction of MA from finger pulse signal collected from 30 human volunteers. A multivariate dataset was generated with systolic peak-aligned Photoplethysmogram (PPG) beats extracted from time series data. After eigenvalues decomposition of the covariance matrix, the original data was reconstructed using the first principal component. The mean correlation coefficient of average beat template of ICA preprocessed data and clean data, averaged over 30 volunteers is 0.9876 while that of PCA preprocessed data with clean data is 0.9778. With white Gaussian noise of known SNR, maximum absolute error for PCA preprocessed data is very small, 3.14% from SNR 25dB onwards. It was also found that beat to beat correlation is higher in the PCA preprocessed data.
Citations
More filters
21 Feb 2007
704 citations
TL;DR: In this article, the authors presented state-of-the-art techniques in both areas of the automated analysis, i.e., motion artifacts removal and heart rate tracking, and concluded that adaptive filtering and multi-resolution decomposition techniques are better for MA removal and machine learning-based approaches are future perspective of heart rate tracker.
Abstract: Non-invasive photoplethysmography (PPG) technology was developed to track heart rate during motion. Automated analysis of PPG has made it useful in both clinical and non-clinical applications. However, PPG-based heart rate tracking is a challenging problem due to motion artifacts (MAs) which are main contributors towards signal degradation as they mask the location of heart rate peak in the spectra. A practical analysis system must have good performance in MA removal as well as in tracking. In this article, we have presented state-of-art techniques in both areas of the automated analysis, i.e., MA removal and heart rate tracking, and have concluded that adaptive filtering and multi-resolution decomposition techniques are better for MA removal and machine learning-based approaches are future perspective of heart rate tracking. Hence, future systems will be composed of machine learning-based trackers fed with either empirically decomposed signal or from output of adaptive filter.
33 citations
References
More filters
TL;DR: Overall, the photoplethysmogram provides a wealth of circulatory information, but its complex etiology may be a limitation in some novel applications.
Abstract: The photoplethysmogram is a noninvasive circulatory signal related to the pulsatile volume of blood in tissue and is displayed by many pulse oximeters and bedside monitors, along with the computed arterial oxygen saturation. The photoplethysmogram is similar in appearance to an arterial blood pressure waveform. Because the former is noninvasive and nearly ubiquitous in hospitals whereas the latter requires invasive measurement, the extraction of circulatory information from the photoplethysmogram has been a popular subject of contemporary research. The photoplethysmogram is a function of the underlying circulation, but the relation is complicated by optical, biomechanical, and physiologic covariates that affect the appearance of the photoplethysmogram. Overall, the photoplethysmogram provides a wealth of circulatory information, but its complex etiology may be a limitation in some novel applications.
345 citations
"Finger photoplethysmogram signal en..." refers background in this paper
...Over the last decade, PPG technology has got significant importance from the medical science and biomedical research community [2-3]....
[...]
TL;DR: An automated, two-stage PPG data processing method to minimize the effects of motion artifacts and presents novel and consistent techniques to detect the presence of motion artifact in PPGs given higher order statistical information present in the data.
Abstract: Corruption of photopleythysmograms (PPGs) by motion artifacts has been a serious obstacle to the reliable use of pulse oximeters for real-time, continuous state-of-health monitoring. In this paper, we propose an automated, two-stage PPG data processing method to minimize the effects of motion artifacts. The technique is based on our prior work related to motion artifact detection (stage 1) [R. Krishnan, B. Natarajan, and S. Warren, ``Analysis and detection of motion artifacts in photoplethysmographic data using higher order statistics,'' in Proc. IEEE Int. Conf. Acoust., Speech, Signal Process. (ICASSP 2008), Las Vegas, Nevada, Apr. 2008, pp. 613-616] and motion artifact reduction (stage 2) [R. Krishnan, B. Natarajan, and S. Warren, ``Motion artifact reduction in photoplethysmography using magnitude-based frequency domain independent component analysis,'' in Proc. 17th Int. Conf. Comput. Commun. Network, St. Thomas, Virgin Islands, Aug. 2008, pp. 1-5]. Regarding stage 1, we present novel and consistent techniques to detect the presence of motion artifact in PPGs given higher order statistical information present in the data. We analyze these data in the time and frequency domains (FDs) and identify metrics to distinguish between clean and motion-corrupted data. A Neyman-Pearson detection rule is formulated for each of the metrics. Furthermore, by treating each of the metrics as observations from independent sensors, we employ hard fusion and soft fusion techniques presented in [Z. Chair and P. Varshney, ``Optimal data fusion in multiple sensor detection systems,'' IEEE Trans. Aerosp. Electron. Syst., AES, vol. 1, no. 1, pp. 98-101, Jan. 1986] and [C. C. Lee and J. J. Chao, ``Optimum local decision space partitioning for distributed detection,'' IEEE Trans. Aerosp. Electron. Syst., AES, vol. 25, no. 7, pp. 536-544, Jul. 1989], respectively, in order to fuse individual decisions into a global system decision. For stage two, we propose a motion artifact reduction method that is effective even in the presence of severe subject movement. The approach involves an enhanced preprocessing unit consisting of a motion detection unit (MDU, developed in this paper), period estimation unit, and Fourier series reconstruction unit. The MDU identifies clean data frames versus those corrupted with motion artifacts. The period estimation unit determines the fundamental frequency of a corrupt frame. The Fourier series reconstruction unit reconstructs the final preprocessed signal by utilizing the spectrum variability of the pulse waveform. Preprocessed data are then fed to a magnitude-based FD independent component analysis unit. This helps reduce motion artifacts present at the frequencies of the reconstruction components. Experimental results are presented to demonstrate the efficacy of the overall motion artifact reduction method.
233 citations
"Finger photoplethysmogram signal en..." refers methods in this paper
...A major focus of medical signal processing research involving PPG has been directed towards PPG enhancement, which includes adaptive filters [8], wavelet decomposition methods [9], cycle to cycle Fourier series analysis [10], independent component analysis (ICA) [11][12], and emperical mode decomposition [13]....
[...]
...In [12] frequency domain ICA was used to remove the motion artifact and cross correlation (highest 0....
[...]
TL;DR: It is concluded that support vector machine-based classification and regression techniques provide effective prediction of arterial stiffness from the simple measurement of the digital volume pulse.
Abstract: Cardiovascular disease (CVD) is currently the biggest single cause of mortality in the developed world, hence, the early detection of its onset is vital for effective prevention therapies. Aortic stiffness as measured by aortic pulse wave velocity (PWV) has been shown to be an independent predictor of CVD, however, the measurement of PWV is complex and time consuming. Recent studies have shown that pulse contour characteristics depend on arterial properties such as arterial stiffness. This paper presents a method for estimating PWV from the digital volume pulse (DVP), a waveform that can be rapidly and simply acquired by measuring the transmission of infra-red light through the finger pulp. PWV and DVP were measured on 461 subjects attending a clinic in South East London. Techniques for extracting features from the DVP contour based on physiology and information theory were compared. Low and high stiffness were defined according to a threshold level of PWV chosen to be 10 m/s. Using a support vector machine-based classifier, it is possible to achieve high overall classification rates on unseen data. Further, the use of support vector regression techniques lead to a direct real-valued estimate of PWV which outperforms previous methods based on multilinear regression. We, therefore, conclude that support vector machine-based classification and regression techniques provide effective prediction of arterial stiffness from the simple measurement of the digital volume pulse. This technique could be usefully employed as a cheap and effective CVD screening technique for use in general practice clinics.
163 citations
TL;DR: Experimental results indicate that the proposed method is insensitive to heart rate variation, introduces negligible error in the processed PPG signals due to the additional processing, preserves all the morphological features of the PPG, provides 35 dB reduction in motion artifacts, and achieves a data compression factor of 12.
Abstract: Pulse oximeters require artifact-free clean photoplethysmograph (PPG) signals obtained at red and infrared (IR) wavelengths for the estimation of the level of oxygen saturation ( SpO2) in the arterial blood of a patient. Movement of a patient corrupts a PPG signal with motion artifacts and introduces large errors in the computation of SpO2. A novel method for removing motion artifacts from corrupted PPG signals by applying Fourier series analysis on a cycle-by-cycle basis is presented in this paper. Aside from artifact reduction, the proposed method also provides data compression. Experimental results indicate that the proposed method is insensitive to heart rate variation, introduces negligible error in the processed PPG signals due to the additional processing, preserves all the morphological features of the PPG, provides 35 dB reduction in motion artifacts, and achieves a data compression factor of 12.
152 citations
"Finger photoplethysmogram signal en..." refers methods in this paper
...A major focus of medical signal processing research involving PPG has been directed towards PPG enhancement, which includes adaptive filters [8], wavelet decomposition methods [9], cycle to cycle Fourier series analysis [10], independent component analysis (ICA) [11][12], and emperical mode decomposition [13]....
[...]
16 May 1998
TL;DR: The ring sensor is worn by the patient at all times, hence the health status is monitored 24 hours a day for a tele-nursing system and any trait of abnormal health status and possible accidents is detected by analyzing the sensor data.
Abstract: This paper presents the development of the ring sensor to monitor a patient 24 hours a day for a tele-nursing system. The ring sensor is worn by the patient at all times, hence the health status is monitored 24 hours a day. The sensors packed into the ring include LEDs with different wavelengths, and technologies of photoplethysmography and pulse oximetry are implemented on the ring. The sensor data are transmitted to a computer through the digital wireless communication link and the patient status is analyzed continually and remotely. Any trait of abnormal health status and possible accidents is detected by analyzing the sensor data. A combination of a global receiver and multiple local ones are used to estimate the patient's location and activity. Both the physiological data and the position information can be used to make an accurate decision as to whether a warning signal must be sent to a medical professional caring the patient. An issue of power reduction for miniaturization of the ring sensor is also addressed.
124 citations