Home
Search
Collections
Journals
About
Contact us
My IOPscience
Development of a noncontact heart rate monitoring system for sedentary behavior based on an accelerometer attached to a chair
This content has been downloaded from IOPscience. Please scroll down to see the full text. 2015 Physiol. Meas. 36 N61 (http://iopscience.iop.org/0967-3334/36/3/N61) View the table of contents for this issue, or go to the journal homepage for more
Download details: IP Address: 169.230.243.252 This content was downloaded on 21/02/2015 at 11:35
Please note that terms and conditions apply.
Institute of Physics and Engineering in Medicine Physiol. Meas. 36 (2015) N61–70
Physiological Measurement doi:10.1088/0967-3334/36/3/N61
Note
Development of a noncontact heart rate monitoring system for sedentary behavior based on an accelerometer attached to a chair Eunho Lim1, Hyo-Ki Lee1, Hyoun-Seok Myoung and Kyoung-Joung Lee2 Department of Biomedical Engineering, Yonsei University, Wonju, Gangwondo, Korea E-mail: [email protected] Received 24 November 2014 Accepted for publication 16 December 2014 Published 16 February 2015 Abstract
Although people spend a third of their day engaged in sedentary activities, research on heart activity during sitting is almost nonexistent because of the discomfort experienced when electrocardiogram (ECG) measurement electrodes are attached to the body. Accordingly, in this study, a system was developed to monitor heart rate (HR) in a noncontact and unconstrained way while subjects were seated, by attaching an accelerometer on the backrest of a chair. Acceleration signals were obtained three times from 20 healthy adults, a detection algorithm was applied, and HR detection performance was evaluated by comparing the R-peak values from the ECG. The system had excellent performance results, with a sensitivity of 96.10% and a positive predictive value of 96.43%. In addition, the HR calculated by the new system developed in this study was compared with HR calculated using ECG. The new system exhibited excellent performance; its results were strongly correlated with that of ECG (r = 0.97, p ≪ 0.0001; average difference of −0.08 ± 4.60 [mean ± 1.96∙standard deviation] in Bland–Altman analysis). Accordingly, the method presented in this study is expected to be applicable for evaluating diverse autonomic nervous system components in a noncontact and unconstrained way using an accelerometer to monitor the HR of sedentary workers or adolescents. Keywords: noncontact heart rate monitoring, sedentary behavior, accelerometer, chair (Some figures may appear in colour only in the online journal) 1 2
These authors contributed equally to this manuscript. Author to whom any correspondence should be addressed.
0967-3334/15/030N61+10$33.00 © 2015 Institute of Physics and Engineering in Medicine Printed in the UK
N61
E Lim et al
Physiol. Meas. 36 (2015) N61
1. Introduction Because workers or adolescents spend more than one third of their average day sitting in a chair in an office or at school, it is important to autonomously monitor individuals’ heart rate (HR) while they are seated. Research has indicated that there is a higher likelihood of developing cardiovascular disease the longer that individuals are seated (Healy et al 2011, Carson et al 2014). HR monitoring allows autonomic neurological system evaluation, which enables cardiovascular disease degree predictions, stress indices evaluation, and renal failure disease anticipation (Acharya et al 2006). A traditional method to monitor HR is to use multiple electrodes and conduct an electrocardiogram (ECG). An ECG can precisely diagnose heart disease, but electrodes should be attached to the body continuously which can cause daily HR monitoring to be uncomfortable. Recently, research has been presented that discusses how to conveniently measure HR during daily life. An electromechanical film (EMFi) sensor has been used as a sitting cushion to obtain individual HR, but it is sensitive to temperature, must be attached to the body, and an accelerometer is also needed to eliminate noise (Postolache et al 2010). A Doppler radar can be used to record HR in a noncontact and unconstrained way, but the results are significantly affected by ambient movements and electronic waves (Suzuki et al 2009a). An optical vibrocardiography can also be used to record HR in a noncontact way, but the equipment is expensive, and a complex optical interface should be used (Kranjec et al 2014). An accelerometer is inexpensive and sufficiently sensitive to measure subtle vibrations in the body; therefore, it is very easy to use to monitor individuals’ HR in daily life. Recent studies have attached an accelerometer or a smart phone with a built in accelerometer to the body to monitor HR (Phan et al 2008, Bryant et al 2010, Kwon et al 2011), but to date, no study has attached an accelerometer to a chair to monitor individuals’ HR during daily life. Therefore, the purpose of this study was to develop a system to monitor individual HR in a noncontact and unconstrained way by attaching an accelerometer to a chair. The HR signals are recorded when vibrations in the body from heart beats are delivered to a chair and sensed by an accelerometer. Our basic premise is that the performance of the HR monitoring system based on an accelerometer attached to a chair is comparable to the HR results calculated from an ECG and to other existing noncontact and unconstrained HR monitoring systems. If this assumption holds true, the HR of people who lead a sedentary lifestyle in their office or school may be simply monitored by the proposed method, which will be also useful for evaluating their autonomic nervous system through HR analysis. 2. Methods 2.1. Subjects
20 healthy adults (age: 25.9 ± 2.7 years old, body mass index: 21.9 ± 1.8 kg m−2) participated in this study. The subjects were informed about the experimental procedure and potential risk factors and consented to take part in this study. 2.2. Experimental system
The entire system to monitor subjects’ HR from acceleration signals was composed of sensor, analogue processing, and digital processing parts, as shown in figure 1. In the sensor part, to obtain heart beat signals in an unconstrained way, a three-axial accelerometer (LIS3L02AL, STMicroelectronics, USA) analogue output type, with a maximum amplitude at ±2 g, was N62
E Lim et al
Physiol. Meas. 36 (2015) N61
Figure 1. Flow diagram for the noncontact HR monitoring system based on an accelerometer attached to a chair.
attached to the backrest of a no-wheel task chair back (PIZZA, Fursys Inc, Republic of Korea), where a semi-synchronized tilting mechanism was applied. The analogue processing part consisted of high and low pass filters, whose bandwidth was between 0.5 and 23.4 Hz, and a 940 times amplitude circuit. In the digital processing part, signals that passed through the analogue processing part were sent to a personal computer at 12 bits and a sampling rate of 200 Hz using a micro controller unit (PIC18F4523, Microchip, USA). The transmitted signals were displayed and stored using LabVIEW (National Instruments Co., USA) and analyzed and evaluated with MATLAB (Mathworks Inc, USA). 2.3. Experimental procedure
The participants sat in the chair which had the designed system attached to it, and acceleration signals were obtained three times from each of the participants. To evaluate the HR extracted from the designed system, ECG signals were synchronized and obtained. The acceleration signals obtained from the system were verified to be affected by the location of the attached sensor and subjects’ posture and movements. It was also verified that the antero-posterior direction (z axis in figure 1) best reflected the HR (Lim et al 2014). The acceleration signals obtained from the backrest of the chair back best reflected heart movements. Heart movements were best reflected when acceleration signals were obtained slightly distant from the chair back without the subjects completely leaning on the chair backrest. Therefore, acceleration signals were obtained at the attachment location and in the posture shown in figure 1. Since the system designed to monitor individuals’ HR measured subtle vibrations of the human body that result from heart movements, the motion artifacts may cause errors when heartbeats are obtained. Therefore, participants were asked to refrain from moving as much as possible. Data were obtained for two minutes and thirty seconds when the subjects were seated and not moving. To validate the obtained data, the data for the two minute period, excluding the first and last 15 s, were compared with the ECG results. 2.4. Algorithm to estimate HR
The antero-posterior signals that best reflected heart movements were used to develop the algorithm (Lim et al 2014). To separate the frequency band related to heartbeats, acceleration signals were passed through a digital low pass filter (cutoff frequency of 8 Hz, 2nd-order Butterworth filter). Figure 2(a) displays signals obtained from the accelerometer attached to the chair, and figure 2(b) shows signals that passed through the filter. To extract HR signals from acceleration signals, all peak and trough points of the filtered acceleration signals were N63
E Lim et al
Physiol. Meas. 36 (2015) N61
Figure 2. Signal processing for detecting HR from an accelerometer signal and the ECG as a reference. (a) a raw accelerometer signal, (b) (pink) stars for peak points and (green) circle points for trough points in the filtered accelerometer signal, (c) the difference between peak and trough point values (dashed line indicates a threshold for decision of point (PK’) in order to calculate HR), (d) (red) circle points for PK’ in the filtered accelerometer signal, (e) (cyan) square for R-peak points in the ECG (ECG) as a reference.
detected (figure 2(b)). To obtain peak points related to HR using the detected peak and trough points, the threshold was determined, as shown in formula (1), using the amplitude of peak points and previous trough points from acceleration signals during the initial three seconds (figure 2(c)). TH = 0.6 ⋅ argmaxm(PK [m ] − TR [m ]) , m = 0, 1, ⋯ , M .
(1)
TH, PK, TR, and M refer to the threshold, amplitude of peak point, amplitude of trough point, and the total number of PKs or TRs that occurred during three seconds, respectively. It was assumed that PK[m] occurred after TR[m]. The coefficient 0.6 was determined experientially and was used to establish the threshold value. The PK with a difference from TR that exceeded TH was selected as the point to calculate HR. If the two PK points that exceeded TH occurred within 90 samples (0.45 s), the largest PK point was selected as a PK point (PKʹ) to calculate HR. This process was repeated to finally detect PKʹ (the red circle in figure 2(d)), and HR was N64
E Lim et al
Physiol. Meas. 36 (2015) N61
calculated using PKʹ. Figures 2(d) and (e) show that PKʹ extracted from acceleration signals corresponded with the R-peak from the ECG. 2.5. Performance evaluation
To evaluate the performance of the method for determining HR using an accelerometer attached to a chair, the sensitivity and the positive predictive value (PPV) to the detected PKʹ were calculated. The PKʹ points extracted from acceleration signals are difficult to precisely evaluate because of their delayed occurrence compared to R-peak values from ECG. Accordingly, this study calculated delay difference between first PKʹ point and first R-peak point and computed sensitivity (Sen) and PPV, as shown in formulas (2) and (3), by taking the window of 30 sample values (0.15 s) after the sample time difference occurred, following the R-peak. Sen (%) =
TP ⋅ 100(%), TP + FN
(2)
TP ⋅ 100(%). TP + FP
(3)
PPV (%) =
TP is a true positive and refers to a case when PKʹ was detected within the window, FN is a false negative and represents a case when PKʹ was not detected in the window, and FP is a false positive and indicates a case when PKʹ was detected outside of the window. Moreover, a correlation analysis was employed to analyze the correlation between instant beats per minute (BPM) extracted from acceleration signals and instant BPM from ECG signals and a Bland– Altman analysis was performed to clearly express the degree of difference. All analyses were performed after median filtering, interpolation and resampling with 1 Hz from both R-peak and PKʹ intervals. 3. Results Table 1 shows the results of sensitivity and PPV when PKʹ was detected from acceleration signals and R-peak values were obtained from ECG, and the average sensitivity and PPV were 96.10% and 96.43%, respectively. Figure 3 illustrates the scatter plot and Pearson’s correlation coefficient for instant BPM from between acceleration signals and ECG signals. The correlation coefficient for the data on the twenty subjects obtained by three repetitive evaluations was 0.97 (p ≪ 0.0001), signifying a statistically significant correlation. Figure 4 shows the Bland–Altman plot for instant BPM from between acceleration signals and ECG signals. The asterisk (*) refers to BPM difference between the PKʹ point and R-peak that were detected every time, and was expressed together with the averaged value (m = −0.08) and 95% limits of agreement (LOA, −4.67 to 4.52). The average of differences was −0.08, and the HR detected by an accelerometer was almost the same as that of an ECG. Values for differences were evenly distributed over average values of HR from the accelerometer and the ECG. 4. Discussion As shown in table 1, the accelerometer HR monitoring system proposed in this study had excellent heartbeat detection performance. Case 15-2 (sensitivity: 86.23%, PPV: 86.23%) had N65
E Lim et al
Physiol. Meas. 36 (2015) N61
Table 1. Heartbeat
detection evaluation (each case number represents the ‘subject number-number of experiments’).
Cases 1-1 1–2 1–3 2-1 2-2 2–3 3-1 3-2 3-3 4-1 4-2 4-3 5-1 5-2 5-3 6-1 6-2 6-3 7-1 7-2 7-3 8-1 8-2 8-3 9-1 9-2 9-3 10–1 10–2 10–3 11-1 11–2 11–3 12-1 12-2 12–3 13-1 13-2 13-3 14-1 14-2 14-3 15-1 15-2 15-3 16-1
R-peaks 149 140 139 136 167 157 154 154 155 135 119 156 169 173 154 136 132 131 149 157 138 156 156 156 151 157 161 122 114 121 162 171 157 128 127 129 167 153 167 117 121 122 155 138 130 199
True positive 147 138 139 136 167 157 154 152 155 132 117 154 168 169 149 136 129 128 143. 152 134 149 151 149 151 156 154 120 111 116 153. 156 150 117 113 119 156 144 152 117 117 119 151 119 114 196
False negative 2 2 0 0 0 0 0 2 0 2 2 2 1 4 5 0 3 3 6 5 4 7 5 7 0 1 7 2 3 5 9 15 7 11 14 10 11 9 15 0 4 3 4 19 16 3
False positive 2 2 0 0 0 0 0 2 0 2 2 2 0 3 4 0 3 3 6 5 4 8 8 6 0 1 7 2 3 5 8 14 6 10 13 11 11 10 15 2 4 2 4 19 13 1
Sensitivity (%)
PPV (%)
98.66 98.57 100 100 100 100 100 98.70 100 98.51 98.32 98.72 99.41 97.69 96.75 100 97.73 97.71 95.97 96.82 97.10 95.51 96.79 95.51 100 99.36 95.65 98.36 97.37 95.87 94.44 91.23 95.54 91.41 88.98 92.25 93.41 94.12 91.02 100 96.69 97.54 97.42 86.23 87.69 98.49
98.66 98.57 100 100 100 100 100 98.70 100 98.51 98.32 98.72 100 98.26 97.39 100 97.73 97.71 95.97 96.82 97.10 94.90 94.97 96.13 100 99.36 95.65 98.36 97.37 95.87 95.03 91.76 96.15 92.13 89.68 91.54 93.41 93.51 91.02 98.32 96.69 98.35 97.42 86.23 89.76 99.49 (Continued )
N66
E Lim et al
Physiol. Meas. 36 (2015) N61
Table 1. (Continued )
Cases
R-peaks
True positive
False negative
False positive
Sensitivity (%)
PPV (%)
16-2 16-3 17-1 17-2 17-3 18-1 18-2 18-3 19-1 19-2 19-3 20–1 20–2 20–3 Total
199 201 146 146 144 147 151 148 155 145 135 168 164 174 8960
192 195 137 142 144 129 144 144 134 126 120 166 160 171 8610
7 6 9 4 0 18 7 4 21 19 15 2 4 3 349
5 4 8 4 1 17 9 5 14 13 12 1 2 1 319
96.48 97.01 93.84 97.26 100 87.76 95.36 97.30 86.45 86.90 88.89 98.81 97.56 98.28 96.10
97.46 97.99 94.48 97.26 99.31 88.36 94.12 96.64 90.54 90.65 90.91 99.40 98.77 99.42 96.43
Figure 3. Correlation analysis of HR estimated from accelerometer signals and HR
estimated with ECG.
the lowest detection performance because it had a section where the overall size of acceleration signals became small relative to other cases, and as a result, PKʹ was not properly detected because its size was smaller than TH. Therefore, when a regular morphology is repeated regardless of the size of acceleration signals, it is necessary to develop an adaptive threshold that can detect PKʹ and an improved system that can detect acceleration signals with N67
E Lim et al
Physiol. Meas. 36 (2015) N61
Figure 4. Bland–Altman plot of HR estimated from accelerometer signals against the ECG signals as a reference. Table 2.
Comparison of the results of this work with other research.
Authors (year)
Sensor/element
Position
Performance
Suzuki (2009)
Doppler radar
Ambient
Chamadiya (2013)
Capacitive sensor
Chung (2007)
Load cell
Watanabe (2005)
Air cushion (pressure sensor) Accelerometer
In front of the backrest of a chair Underneath the legs of a bed Under the mattress in a bed Behind the backrest of a chair
Correlation coefficient: 0.76 Accuracy: 96.67
This work
Correlation coefficient: > 0.9 Accuracy: 88.17 Sensitivity: 96.10 PPV: 96.43 Correlation coefficient: 0.97
a certain form. The correlation and Bland–Altman analyses results indicated that the outcome of the HR calculations by the monitoring system developed in this study were not statistically different from ECG outcomes. Table 2 displays the comparison between the performance of the existing unconstrained and noncontact HR monitoring system and that of the system developed in this study. The existing HR monitoring system has excellent periodical heartbeat performance but is sensitive to ambient movements (Suzuki et al 2009b) and requires its sensor to be attached to the body (Chamadiya et al 2013). Prior research utilizing a load cell or an air cushion for an unconstrained and noncontact method has only been applied to a bed, not N68
E Lim et al
Physiol. Meas. 36 (2015) N61
to a chair (Watanabe et al 2005, Chung et al 2007). In the present study, we attached a small accelerometer to a chair to develop an HR monitoring system that exhibited the equivalent performance outcomes as existing HR monitoring systems. To the best of our knowledge, this was the first study to investigate the feasibility of an accelerometer attached behind the backrest of a chair to monitor individuals’ HR during sedentary behavior. The system developed in this study was able to detect heartbeats by attaching a low-cost accelerometer to a task chair used in ordinary life and therefore enables unconstrained and noncontact HR monitoring in offices and schools. However, because we attached the accelerometer only to a chair without wheels that had a semi-synchronized tilting mechanism, additional research to precisely detect an individual’s HR regardless of the chair type is necessary so that subject’s HR may be precisely detected from various types of chairs. In addition, the fixed threshold used to extract HR from an accelerometer is disadvantageous because of increasing the number of false negatives when body movements occur. Therefore, it is necessary to develop an improved algorithm that may precisely extract heartbeats regardless of body movements. Recent research has been presented that discusses how to remove the motion artifact (Raya et al 2002, Gibbs and Asada 2005, Postolache et al 2010). Our future research should develop an improved algorithm that may precisely extract heartbeats regardless of body movements by using an adaptive noise filter. In addition, because the findings were derived after an accelerometer was attached to the backrest of a chair back and the subjects did not lean on the backrest completely, additional research is necessary for accurate HR detection for various sensor attachment locations and postures. Future research should involve a larger number of participants to evaluate HR monitoring performance. 5. Conclusion In this paper, we proposed and evaluated a method to measure HR signals in a noncontact and unconstrained way using an acceleration system that was attached to a chair while subjects are seated. The method presented in this paper allows HR to be extracted in a noncontact and unconstrained way and therefore will be useful for evaluating the autonomic nervous system of individuals that sit in chairs in offices or classrooms. This study only conducted HR monitoring, but if an algorithm to monitor respiration is also developed, it could lead to development of a cardiopulmonary function monitoring system for ubiquitous healthcare. Acknowledgments Followings are results of a study on the ‘Leaders Industry-University Cooperation (LINC)’ Project (2014-D-7266–010100) supported by the Ministry of Education (MOE). And this work was supported in part by the Yonsei University Research Fund of 2014. References Acharya U R, Joseph K P, Kannathal N, Lim C M and Suri J S 2006 Heart rate variability: a review Med. Biol. Eng. Comput. 44 1031–51 Bryant D, Ravindran S, Magotra N and Northrup S 2010 Real-time implementation of a chest-worn accelerometer based heart monitoring system 53rd IEEE Int. Midwest Symp. on Circuits and Systems (Seattle, WA, 2010) pp 1057–60 N69
E Lim et al
Physiol. Meas. 36 (2015) N61
Carson V, Staiano A and Katzmarzyk P 2014 Physical activity, screen time, and sitting among US adolescents Pediatr. Exerc. Sci. (at press) Chamadiya B, Mankodiya K, Wagner M and Hofmann U G 2013 Textile-based, contactless ECG monitoring for non-ICU clinical settings J. Ambient Intell. Human. Comput. 4 791–800 Chung G S, Choi B H, Jeong D U and Park K S 2007 Noninvasive heart rate variability analysis using loadcell-installed bed during sleep EMBS 2007: Proc. 29th Annual Int. Conf. of IEEE Engineering in Medicine and Biology Society (Lyon, France) pp 2357–60 Gibbs P and Asada H H 2005 Reducing motion artifact in wearable bio-sensors using MEMS accelerometers for active noise cancellation Proc. American Control Conf. 2005 (Portland, OR USA) pp 1581–6 Healy G N, Matthews C E, Dunstan D W, Winkler E A H and Owen N 2011 Sedentary time and cardiometabolic biomarkers in US adults: NHANES 2003–06 Eur. Heart J. 32 590–7 Kranjec J, Beguš S, Geršak G and Drnovšek J 2014 Non-contact heart rate and heart rate variability measurements: a review Biomed. Signal Process. Control 13 102–12 Kwon S J, Lee J S, Chung G S and Park K S 2011 Validation of heart rate extraction through an iPhone accelerometer EMBS 2011: Proc. 33rd Annual Int. Conf. of IEEE Engineering in Medicine and Biology Society (Boston, MA) pp 5260–3 Raya M A D and Sison L G 2002 Adaptive noise cancelling of motion artifact in stress ECG signals using accelerometer EMBS 2002: Proc. 24th Annual Int. Conf. of IEEE Engineering in Medicine and Biology Society (Texas, USA) pp 1756–7 Lim E, Lee H-K, Myoung H-S and Lee K-J 2014 Preliminary study for evaluation of accelerometric system attached on chair for unconstrained heart rate monitoring Int. Biomedical Engineering Conf. 2014 (Gwangju, Republic of Korea) 717 Phan D H, Bonnet S, Guillemaud R, Castelli E and Pham Thi N Y 2008 Estimation of respiratory waveform and heart rate using an accelerometer EMBS 2008: Proc. 30th Annual Int. Conf. of IEEE Engineering in Medicine and Biology Society (Vancouver, Canada) pp 4916–9 Postolache O A, Girao P S, Mendes J, Pinheiro E C and Postolache G 2010 Physiological parameters measurement based on wheelchair embedded sensors and advanced signal processing IEEE Trans. Instrum. Meas. 59 2564–74 Suzuki S, Matsui T, Gotoh S, Mori Y and Takase B 2009a Development of non-contact monitoring system of heart rate variability (HRV)—an approach of remote sensing for ubiquitous technology EHAWC 2009: Proc. Int. Conf. on Ergonomics and Health Aspects of Work with Computers: Held as Part of HCI Int. 2009 (San Diego, CA USA) pp 195–203 Suzuki S, Matsui T, Kawahara H, Ichiki H and Shimizu J 2009b A non-contact vital sign monitoring system for ambulances using dual-frequency microwave radars Med. Biol. Eng. Comput. 47 101–5 Watanabe K, Watanabe T, Watanabe H, Ando H, Ishikawa T and Kobayashi K 2005 Noninvasive measurement of heartbeat, respiration, snoring and body movements of a subject in bed via a pneumatic method IEEE Trans. Biomed. Eng. 52 2100–7
N70
Search
Collections
Journals
About
Contact us
My IOPscience
Development of a noncontact heart rate monitoring system for sedentary behavior based on an accelerometer attached to a chair
This content has been downloaded from IOPscience. Please scroll down to see the full text. 2015 Physiol. Meas. 36 N61 (http://iopscience.iop.org/0967-3334/36/3/N61) View the table of contents for this issue, or go to the journal homepage for more
Download details: IP Address: 169.230.243.252 This content was downloaded on 21/02/2015 at 11:35
Please note that terms and conditions apply.
Institute of Physics and Engineering in Medicine Physiol. Meas. 36 (2015) N61–70
Physiological Measurement doi:10.1088/0967-3334/36/3/N61
Note
Development of a noncontact heart rate monitoring system for sedentary behavior based on an accelerometer attached to a chair Eunho Lim1, Hyo-Ki Lee1, Hyoun-Seok Myoung and Kyoung-Joung Lee2 Department of Biomedical Engineering, Yonsei University, Wonju, Gangwondo, Korea E-mail: [email protected] Received 24 November 2014 Accepted for publication 16 December 2014 Published 16 February 2015 Abstract
Although people spend a third of their day engaged in sedentary activities, research on heart activity during sitting is almost nonexistent because of the discomfort experienced when electrocardiogram (ECG) measurement electrodes are attached to the body. Accordingly, in this study, a system was developed to monitor heart rate (HR) in a noncontact and unconstrained way while subjects were seated, by attaching an accelerometer on the backrest of a chair. Acceleration signals were obtained three times from 20 healthy adults, a detection algorithm was applied, and HR detection performance was evaluated by comparing the R-peak values from the ECG. The system had excellent performance results, with a sensitivity of 96.10% and a positive predictive value of 96.43%. In addition, the HR calculated by the new system developed in this study was compared with HR calculated using ECG. The new system exhibited excellent performance; its results were strongly correlated with that of ECG (r = 0.97, p ≪ 0.0001; average difference of −0.08 ± 4.60 [mean ± 1.96∙standard deviation] in Bland–Altman analysis). Accordingly, the method presented in this study is expected to be applicable for evaluating diverse autonomic nervous system components in a noncontact and unconstrained way using an accelerometer to monitor the HR of sedentary workers or adolescents. Keywords: noncontact heart rate monitoring, sedentary behavior, accelerometer, chair (Some figures may appear in colour only in the online journal) 1 2
These authors contributed equally to this manuscript. Author to whom any correspondence should be addressed.
0967-3334/15/030N61+10$33.00 © 2015 Institute of Physics and Engineering in Medicine Printed in the UK
N61
E Lim et al
Physiol. Meas. 36 (2015) N61
1. Introduction Because workers or adolescents spend more than one third of their average day sitting in a chair in an office or at school, it is important to autonomously monitor individuals’ heart rate (HR) while they are seated. Research has indicated that there is a higher likelihood of developing cardiovascular disease the longer that individuals are seated (Healy et al 2011, Carson et al 2014). HR monitoring allows autonomic neurological system evaluation, which enables cardiovascular disease degree predictions, stress indices evaluation, and renal failure disease anticipation (Acharya et al 2006). A traditional method to monitor HR is to use multiple electrodes and conduct an electrocardiogram (ECG). An ECG can precisely diagnose heart disease, but electrodes should be attached to the body continuously which can cause daily HR monitoring to be uncomfortable. Recently, research has been presented that discusses how to conveniently measure HR during daily life. An electromechanical film (EMFi) sensor has been used as a sitting cushion to obtain individual HR, but it is sensitive to temperature, must be attached to the body, and an accelerometer is also needed to eliminate noise (Postolache et al 2010). A Doppler radar can be used to record HR in a noncontact and unconstrained way, but the results are significantly affected by ambient movements and electronic waves (Suzuki et al 2009a). An optical vibrocardiography can also be used to record HR in a noncontact way, but the equipment is expensive, and a complex optical interface should be used (Kranjec et al 2014). An accelerometer is inexpensive and sufficiently sensitive to measure subtle vibrations in the body; therefore, it is very easy to use to monitor individuals’ HR in daily life. Recent studies have attached an accelerometer or a smart phone with a built in accelerometer to the body to monitor HR (Phan et al 2008, Bryant et al 2010, Kwon et al 2011), but to date, no study has attached an accelerometer to a chair to monitor individuals’ HR during daily life. Therefore, the purpose of this study was to develop a system to monitor individual HR in a noncontact and unconstrained way by attaching an accelerometer to a chair. The HR signals are recorded when vibrations in the body from heart beats are delivered to a chair and sensed by an accelerometer. Our basic premise is that the performance of the HR monitoring system based on an accelerometer attached to a chair is comparable to the HR results calculated from an ECG and to other existing noncontact and unconstrained HR monitoring systems. If this assumption holds true, the HR of people who lead a sedentary lifestyle in their office or school may be simply monitored by the proposed method, which will be also useful for evaluating their autonomic nervous system through HR analysis. 2. Methods 2.1. Subjects
20 healthy adults (age: 25.9 ± 2.7 years old, body mass index: 21.9 ± 1.8 kg m−2) participated in this study. The subjects were informed about the experimental procedure and potential risk factors and consented to take part in this study. 2.2. Experimental system
The entire system to monitor subjects’ HR from acceleration signals was composed of sensor, analogue processing, and digital processing parts, as shown in figure 1. In the sensor part, to obtain heart beat signals in an unconstrained way, a three-axial accelerometer (LIS3L02AL, STMicroelectronics, USA) analogue output type, with a maximum amplitude at ±2 g, was N62
E Lim et al
Physiol. Meas. 36 (2015) N61
Figure 1. Flow diagram for the noncontact HR monitoring system based on an accelerometer attached to a chair.
attached to the backrest of a no-wheel task chair back (PIZZA, Fursys Inc, Republic of Korea), where a semi-synchronized tilting mechanism was applied. The analogue processing part consisted of high and low pass filters, whose bandwidth was between 0.5 and 23.4 Hz, and a 940 times amplitude circuit. In the digital processing part, signals that passed through the analogue processing part were sent to a personal computer at 12 bits and a sampling rate of 200 Hz using a micro controller unit (PIC18F4523, Microchip, USA). The transmitted signals were displayed and stored using LabVIEW (National Instruments Co., USA) and analyzed and evaluated with MATLAB (Mathworks Inc, USA). 2.3. Experimental procedure
The participants sat in the chair which had the designed system attached to it, and acceleration signals were obtained three times from each of the participants. To evaluate the HR extracted from the designed system, ECG signals were synchronized and obtained. The acceleration signals obtained from the system were verified to be affected by the location of the attached sensor and subjects’ posture and movements. It was also verified that the antero-posterior direction (z axis in figure 1) best reflected the HR (Lim et al 2014). The acceleration signals obtained from the backrest of the chair back best reflected heart movements. Heart movements were best reflected when acceleration signals were obtained slightly distant from the chair back without the subjects completely leaning on the chair backrest. Therefore, acceleration signals were obtained at the attachment location and in the posture shown in figure 1. Since the system designed to monitor individuals’ HR measured subtle vibrations of the human body that result from heart movements, the motion artifacts may cause errors when heartbeats are obtained. Therefore, participants were asked to refrain from moving as much as possible. Data were obtained for two minutes and thirty seconds when the subjects were seated and not moving. To validate the obtained data, the data for the two minute period, excluding the first and last 15 s, were compared with the ECG results. 2.4. Algorithm to estimate HR
The antero-posterior signals that best reflected heart movements were used to develop the algorithm (Lim et al 2014). To separate the frequency band related to heartbeats, acceleration signals were passed through a digital low pass filter (cutoff frequency of 8 Hz, 2nd-order Butterworth filter). Figure 2(a) displays signals obtained from the accelerometer attached to the chair, and figure 2(b) shows signals that passed through the filter. To extract HR signals from acceleration signals, all peak and trough points of the filtered acceleration signals were N63
E Lim et al
Physiol. Meas. 36 (2015) N61
Figure 2. Signal processing for detecting HR from an accelerometer signal and the ECG as a reference. (a) a raw accelerometer signal, (b) (pink) stars for peak points and (green) circle points for trough points in the filtered accelerometer signal, (c) the difference between peak and trough point values (dashed line indicates a threshold for decision of point (PK’) in order to calculate HR), (d) (red) circle points for PK’ in the filtered accelerometer signal, (e) (cyan) square for R-peak points in the ECG (ECG) as a reference.
detected (figure 2(b)). To obtain peak points related to HR using the detected peak and trough points, the threshold was determined, as shown in formula (1), using the amplitude of peak points and previous trough points from acceleration signals during the initial three seconds (figure 2(c)). TH = 0.6 ⋅ argmaxm(PK [m ] − TR [m ]) , m = 0, 1, ⋯ , M .
(1)
TH, PK, TR, and M refer to the threshold, amplitude of peak point, amplitude of trough point, and the total number of PKs or TRs that occurred during three seconds, respectively. It was assumed that PK[m] occurred after TR[m]. The coefficient 0.6 was determined experientially and was used to establish the threshold value. The PK with a difference from TR that exceeded TH was selected as the point to calculate HR. If the two PK points that exceeded TH occurred within 90 samples (0.45 s), the largest PK point was selected as a PK point (PKʹ) to calculate HR. This process was repeated to finally detect PKʹ (the red circle in figure 2(d)), and HR was N64
E Lim et al
Physiol. Meas. 36 (2015) N61
calculated using PKʹ. Figures 2(d) and (e) show that PKʹ extracted from acceleration signals corresponded with the R-peak from the ECG. 2.5. Performance evaluation
To evaluate the performance of the method for determining HR using an accelerometer attached to a chair, the sensitivity and the positive predictive value (PPV) to the detected PKʹ were calculated. The PKʹ points extracted from acceleration signals are difficult to precisely evaluate because of their delayed occurrence compared to R-peak values from ECG. Accordingly, this study calculated delay difference between first PKʹ point and first R-peak point and computed sensitivity (Sen) and PPV, as shown in formulas (2) and (3), by taking the window of 30 sample values (0.15 s) after the sample time difference occurred, following the R-peak. Sen (%) =
TP ⋅ 100(%), TP + FN
(2)
TP ⋅ 100(%). TP + FP
(3)
PPV (%) =
TP is a true positive and refers to a case when PKʹ was detected within the window, FN is a false negative and represents a case when PKʹ was not detected in the window, and FP is a false positive and indicates a case when PKʹ was detected outside of the window. Moreover, a correlation analysis was employed to analyze the correlation between instant beats per minute (BPM) extracted from acceleration signals and instant BPM from ECG signals and a Bland– Altman analysis was performed to clearly express the degree of difference. All analyses were performed after median filtering, interpolation and resampling with 1 Hz from both R-peak and PKʹ intervals. 3. Results Table 1 shows the results of sensitivity and PPV when PKʹ was detected from acceleration signals and R-peak values were obtained from ECG, and the average sensitivity and PPV were 96.10% and 96.43%, respectively. Figure 3 illustrates the scatter plot and Pearson’s correlation coefficient for instant BPM from between acceleration signals and ECG signals. The correlation coefficient for the data on the twenty subjects obtained by three repetitive evaluations was 0.97 (p ≪ 0.0001), signifying a statistically significant correlation. Figure 4 shows the Bland–Altman plot for instant BPM from between acceleration signals and ECG signals. The asterisk (*) refers to BPM difference between the PKʹ point and R-peak that were detected every time, and was expressed together with the averaged value (m = −0.08) and 95% limits of agreement (LOA, −4.67 to 4.52). The average of differences was −0.08, and the HR detected by an accelerometer was almost the same as that of an ECG. Values for differences were evenly distributed over average values of HR from the accelerometer and the ECG. 4. Discussion As shown in table 1, the accelerometer HR monitoring system proposed in this study had excellent heartbeat detection performance. Case 15-2 (sensitivity: 86.23%, PPV: 86.23%) had N65
E Lim et al
Physiol. Meas. 36 (2015) N61
Table 1. Heartbeat
detection evaluation (each case number represents the ‘subject number-number of experiments’).
Cases 1-1 1–2 1–3 2-1 2-2 2–3 3-1 3-2 3-3 4-1 4-2 4-3 5-1 5-2 5-3 6-1 6-2 6-3 7-1 7-2 7-3 8-1 8-2 8-3 9-1 9-2 9-3 10–1 10–2 10–3 11-1 11–2 11–3 12-1 12-2 12–3 13-1 13-2 13-3 14-1 14-2 14-3 15-1 15-2 15-3 16-1
R-peaks 149 140 139 136 167 157 154 154 155 135 119 156 169 173 154 136 132 131 149 157 138 156 156 156 151 157 161 122 114 121 162 171 157 128 127 129 167 153 167 117 121 122 155 138 130 199
True positive 147 138 139 136 167 157 154 152 155 132 117 154 168 169 149 136 129 128 143. 152 134 149 151 149 151 156 154 120 111 116 153. 156 150 117 113 119 156 144 152 117 117 119 151 119 114 196
False negative 2 2 0 0 0 0 0 2 0 2 2 2 1 4 5 0 3 3 6 5 4 7 5 7 0 1 7 2 3 5 9 15 7 11 14 10 11 9 15 0 4 3 4 19 16 3
False positive 2 2 0 0 0 0 0 2 0 2 2 2 0 3 4 0 3 3 6 5 4 8 8 6 0 1 7 2 3 5 8 14 6 10 13 11 11 10 15 2 4 2 4 19 13 1
Sensitivity (%)
PPV (%)
98.66 98.57 100 100 100 100 100 98.70 100 98.51 98.32 98.72 99.41 97.69 96.75 100 97.73 97.71 95.97 96.82 97.10 95.51 96.79 95.51 100 99.36 95.65 98.36 97.37 95.87 94.44 91.23 95.54 91.41 88.98 92.25 93.41 94.12 91.02 100 96.69 97.54 97.42 86.23 87.69 98.49
98.66 98.57 100 100 100 100 100 98.70 100 98.51 98.32 98.72 100 98.26 97.39 100 97.73 97.71 95.97 96.82 97.10 94.90 94.97 96.13 100 99.36 95.65 98.36 97.37 95.87 95.03 91.76 96.15 92.13 89.68 91.54 93.41 93.51 91.02 98.32 96.69 98.35 97.42 86.23 89.76 99.49 (Continued )
N66
E Lim et al
Physiol. Meas. 36 (2015) N61
Table 1. (Continued )
Cases
R-peaks
True positive
False negative
False positive
Sensitivity (%)
PPV (%)
16-2 16-3 17-1 17-2 17-3 18-1 18-2 18-3 19-1 19-2 19-3 20–1 20–2 20–3 Total
199 201 146 146 144 147 151 148 155 145 135 168 164 174 8960
192 195 137 142 144 129 144 144 134 126 120 166 160 171 8610
7 6 9 4 0 18 7 4 21 19 15 2 4 3 349
5 4 8 4 1 17 9 5 14 13 12 1 2 1 319
96.48 97.01 93.84 97.26 100 87.76 95.36 97.30 86.45 86.90 88.89 98.81 97.56 98.28 96.10
97.46 97.99 94.48 97.26 99.31 88.36 94.12 96.64 90.54 90.65 90.91 99.40 98.77 99.42 96.43
Figure 3. Correlation analysis of HR estimated from accelerometer signals and HR
estimated with ECG.
the lowest detection performance because it had a section where the overall size of acceleration signals became small relative to other cases, and as a result, PKʹ was not properly detected because its size was smaller than TH. Therefore, when a regular morphology is repeated regardless of the size of acceleration signals, it is necessary to develop an adaptive threshold that can detect PKʹ and an improved system that can detect acceleration signals with N67
E Lim et al
Physiol. Meas. 36 (2015) N61
Figure 4. Bland–Altman plot of HR estimated from accelerometer signals against the ECG signals as a reference. Table 2.
Comparison of the results of this work with other research.
Authors (year)
Sensor/element
Position
Performance
Suzuki (2009)
Doppler radar
Ambient
Chamadiya (2013)
Capacitive sensor
Chung (2007)
Load cell
Watanabe (2005)
Air cushion (pressure sensor) Accelerometer
In front of the backrest of a chair Underneath the legs of a bed Under the mattress in a bed Behind the backrest of a chair
Correlation coefficient: 0.76 Accuracy: 96.67
This work
Correlation coefficient: > 0.9 Accuracy: 88.17 Sensitivity: 96.10 PPV: 96.43 Correlation coefficient: 0.97
a certain form. The correlation and Bland–Altman analyses results indicated that the outcome of the HR calculations by the monitoring system developed in this study were not statistically different from ECG outcomes. Table 2 displays the comparison between the performance of the existing unconstrained and noncontact HR monitoring system and that of the system developed in this study. The existing HR monitoring system has excellent periodical heartbeat performance but is sensitive to ambient movements (Suzuki et al 2009b) and requires its sensor to be attached to the body (Chamadiya et al 2013). Prior research utilizing a load cell or an air cushion for an unconstrained and noncontact method has only been applied to a bed, not N68
E Lim et al
Physiol. Meas. 36 (2015) N61
to a chair (Watanabe et al 2005, Chung et al 2007). In the present study, we attached a small accelerometer to a chair to develop an HR monitoring system that exhibited the equivalent performance outcomes as existing HR monitoring systems. To the best of our knowledge, this was the first study to investigate the feasibility of an accelerometer attached behind the backrest of a chair to monitor individuals’ HR during sedentary behavior. The system developed in this study was able to detect heartbeats by attaching a low-cost accelerometer to a task chair used in ordinary life and therefore enables unconstrained and noncontact HR monitoring in offices and schools. However, because we attached the accelerometer only to a chair without wheels that had a semi-synchronized tilting mechanism, additional research to precisely detect an individual’s HR regardless of the chair type is necessary so that subject’s HR may be precisely detected from various types of chairs. In addition, the fixed threshold used to extract HR from an accelerometer is disadvantageous because of increasing the number of false negatives when body movements occur. Therefore, it is necessary to develop an improved algorithm that may precisely extract heartbeats regardless of body movements. Recent research has been presented that discusses how to remove the motion artifact (Raya et al 2002, Gibbs and Asada 2005, Postolache et al 2010). Our future research should develop an improved algorithm that may precisely extract heartbeats regardless of body movements by using an adaptive noise filter. In addition, because the findings were derived after an accelerometer was attached to the backrest of a chair back and the subjects did not lean on the backrest completely, additional research is necessary for accurate HR detection for various sensor attachment locations and postures. Future research should involve a larger number of participants to evaluate HR monitoring performance. 5. Conclusion In this paper, we proposed and evaluated a method to measure HR signals in a noncontact and unconstrained way using an acceleration system that was attached to a chair while subjects are seated. The method presented in this paper allows HR to be extracted in a noncontact and unconstrained way and therefore will be useful for evaluating the autonomic nervous system of individuals that sit in chairs in offices or classrooms. This study only conducted HR monitoring, but if an algorithm to monitor respiration is also developed, it could lead to development of a cardiopulmonary function monitoring system for ubiquitous healthcare. Acknowledgments Followings are results of a study on the ‘Leaders Industry-University Cooperation (LINC)’ Project (2014-D-7266–010100) supported by the Ministry of Education (MOE). And this work was supported in part by the Yonsei University Research Fund of 2014. References Acharya U R, Joseph K P, Kannathal N, Lim C M and Suri J S 2006 Heart rate variability: a review Med. Biol. Eng. Comput. 44 1031–51 Bryant D, Ravindran S, Magotra N and Northrup S 2010 Real-time implementation of a chest-worn accelerometer based heart monitoring system 53rd IEEE Int. Midwest Symp. on Circuits and Systems (Seattle, WA, 2010) pp 1057–60 N69
E Lim et al
Physiol. Meas. 36 (2015) N61
Carson V, Staiano A and Katzmarzyk P 2014 Physical activity, screen time, and sitting among US adolescents Pediatr. Exerc. Sci. (at press) Chamadiya B, Mankodiya K, Wagner M and Hofmann U G 2013 Textile-based, contactless ECG monitoring for non-ICU clinical settings J. Ambient Intell. Human. Comput. 4 791–800 Chung G S, Choi B H, Jeong D U and Park K S 2007 Noninvasive heart rate variability analysis using loadcell-installed bed during sleep EMBS 2007: Proc. 29th Annual Int. Conf. of IEEE Engineering in Medicine and Biology Society (Lyon, France) pp 2357–60 Gibbs P and Asada H H 2005 Reducing motion artifact in wearable bio-sensors using MEMS accelerometers for active noise cancellation Proc. American Control Conf. 2005 (Portland, OR USA) pp 1581–6 Healy G N, Matthews C E, Dunstan D W, Winkler E A H and Owen N 2011 Sedentary time and cardiometabolic biomarkers in US adults: NHANES 2003–06 Eur. Heart J. 32 590–7 Kranjec J, Beguš S, Geršak G and Drnovšek J 2014 Non-contact heart rate and heart rate variability measurements: a review Biomed. Signal Process. Control 13 102–12 Kwon S J, Lee J S, Chung G S and Park K S 2011 Validation of heart rate extraction through an iPhone accelerometer EMBS 2011: Proc. 33rd Annual Int. Conf. of IEEE Engineering in Medicine and Biology Society (Boston, MA) pp 5260–3 Raya M A D and Sison L G 2002 Adaptive noise cancelling of motion artifact in stress ECG signals using accelerometer EMBS 2002: Proc. 24th Annual Int. Conf. of IEEE Engineering in Medicine and Biology Society (Texas, USA) pp 1756–7 Lim E, Lee H-K, Myoung H-S and Lee K-J 2014 Preliminary study for evaluation of accelerometric system attached on chair for unconstrained heart rate monitoring Int. Biomedical Engineering Conf. 2014 (Gwangju, Republic of Korea) 717 Phan D H, Bonnet S, Guillemaud R, Castelli E and Pham Thi N Y 2008 Estimation of respiratory waveform and heart rate using an accelerometer EMBS 2008: Proc. 30th Annual Int. Conf. of IEEE Engineering in Medicine and Biology Society (Vancouver, Canada) pp 4916–9 Postolache O A, Girao P S, Mendes J, Pinheiro E C and Postolache G 2010 Physiological parameters measurement based on wheelchair embedded sensors and advanced signal processing IEEE Trans. Instrum. Meas. 59 2564–74 Suzuki S, Matsui T, Gotoh S, Mori Y and Takase B 2009a Development of non-contact monitoring system of heart rate variability (HRV)—an approach of remote sensing for ubiquitous technology EHAWC 2009: Proc. Int. Conf. on Ergonomics and Health Aspects of Work with Computers: Held as Part of HCI Int. 2009 (San Diego, CA USA) pp 195–203 Suzuki S, Matsui T, Kawahara H, Ichiki H and Shimizu J 2009b A non-contact vital sign monitoring system for ambulances using dual-frequency microwave radars Med. Biol. Eng. Comput. 47 101–5 Watanabe K, Watanabe T, Watanabe H, Ando H, Ishikawa T and Kobayashi K 2005 Noninvasive measurement of heartbeat, respiration, snoring and body movements of a subject in bed via a pneumatic method IEEE Trans. Biomed. Eng. 52 2100–7
N70
