IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 7, JULY 2014
2125
Unconstrained Sleep Apnea Monitoring Using Polyvinylidene Fluoride Film-Based Sensor Su Hwan Hwang, Hong Ji Lee, Hee Nam Yoon, Da Woon Jung, Yu-Jin G. Lee, Yu Jin Lee, Do-Un Jeong, and Kwang Suk Park∗ , Senior Member, IEEE
Abstract—We established and tested an unconstrained sleep apnea monitoring method using a polyvinylidene (PVDF) filmbased sensor for continuous and accurate monitoring of apneic events occurred during sleep. Twenty-six sleep apnea patients and six normal subjects participated in this study. Subjects’ respiratory signals were measured using the PVDF-based sensor during polysomnography. The PVDF sensor comprised a 4 × 1 array, and a thin silicon pad was placed over the sensor to prevent damage. Total thickness of the merged system was approximately 1.1 mm which was thin enough to prevent the subject from being consciously aware of its presence. It was designed to be placed under subjects’ backs and installed between a bed cover and mattress. The proposed method was based on the standard deviation of the PVDF signals, and it was applied to a test set for detecting apneic events. The method’s performance was assessed by comparing the results with a sleep physician’s manual scoring. The correlation coefficient for the apnea-hypopnea index (AHI) values between the methods was 0.94 (p < 0.001). The areas under the receiver operating curves at three AHI threshold levels (>5, >15, and >20) for sleep apnea diagnosis were 0.98, 0.99, and 0.98, respectively. For min-by-min apnea detection, the method classified sleep apnea with an average sensitivity of 72.9%, specificity of 90.6%, accuracy of 85.5%, and kappa statistic of 0.60. The developed system and method can be applied to sleep apnea detection in home or ambulatory monitoring. Index Terms—Apnea-hypopnea index (AHI), PVDF-sensor, unconstrained sleep apnea monitoring.
I. INTRODUCTION LEEP APNEA is a typical sleep-related breathing disorder (SRBD) characterized by frequent, abnormal cessation of respiration during sleep [1]. During the apneic period, there is an increased effort in breathing, leading to arousal and sleep fragmentation [2]. Thus, severe and frequent sleep apnea disrupts
S
Manuscript received July 9, 2013; revised November 4, 2013; accepted March 25, 2014. Date of publication April 4, 2014; date of current version June 14, 2014. This work was supported by Samsung Electronics, Inc. Asterisk indicates corresponding author. S. H. Hwang, H. J. Lee, H. N. Yoon, and D. W. Jung are with the Interdisciplinary Program in Bioengineering, Seoul National University, Seoul 110-799, Korea (e-mail: [email protected]; [email protected]; [email protected]; [email protected]). Y.-J. G. Lee is with the Seoul Metropolitan Eunpyeong Hospital, Seoul 122-913, Korea (e-mail: [email protected]). Y. J. Lee and D.-U. Jeong are with the Department of Neuropsychiatry and the Center for Sleep and Chronobiology, Seoul National University Hospital, Seoul 110-744, Korea (e-mail: [email protected]; [email protected]). ∗ K. S. Park is with the Department of Biomedical Engineering, College of Medicine, Seoul National University, Seoul 110-799, Korea (e-mail: [email protected]). Digital Object Identifier 10.1109/TBME.2014.2314452
the sleep architecture of subjects and can lead to sleep disorders such as severe snoring [3], fatigue, daytime sleepiness [4], and systemic hypertension [5]. In addition, apnea-induced hypoxia during sleep can cause stroke [6], arrhythmias [7], diabetes [8], and cardiovascular diseases [9]. In previous study, Young et al. reported that approximately 2% of adult women and 4% of adult men in the middle-age range are affected by sleep apnea [10]. Kim et al. analyzed data to determine the prevalence of sleep-disordered breathing (SDB) in 457 Korean adults aged 40–69 years and 27% of men and 16% of women, respectively, had SDB [11]. Furthermore, a recent study of SDB in adults discovered the actual prevalence rates of SDB representing substantial increase over the last two decades (10% among 30–49-year-old men; 17% among 50–70-year-old men; 3% among 30–49-year-old women; and 9% among 50–70-yearold women) [12]. In a general sleep and sleep-related disorder monitoring system, polysomnography (PSG) has been regarded as the goldstandard method. According to the PSG recording and diagnosis rules, apneic events are detected based on the nasal-oral airflow amplitude and blood oxygen saturation (SaO2 ) level [13]. Even though PSG has been used for the assessment of sleep apnea, the detection of apneic events using PSG has some problems. First, PSG recording during sleep is an inconvenient experience for subjects because numerous sensors are attached to the patient’s body (SpO2 ) and face (nasal-oral airflow). Second, the manual scoring of nocturnal apneic events from PSG data is a very time-consuming and laborious process, and it demands specially trained sleep experts. For overcoming some of these drawbacks, in recent years, many alternative methods without using PSG to detect apneic or hypopneic events have been proposed. These studies have involved sleep apnea detection based on several biosignals. For example, electrocardiogram (ECG)-based studies have shown that RR-interval- or R-peak amplitude-based methods [14]–[17] are useful for apnea detection. A ballistocardiogram (BCG)-based system measured the respiration rate with high correlation using an air mattress with a balancing tube [18]. Pulse oximetry- and respiratory-based studies [19], [20] revealed a relatively high correlation coefficient (r > 0.9) between the apnea-hypopnea index (AHI) from PSG and the suggested one. Despite these efforts, a clearly superior method or system for unconstrained sleep apnea monitoring still does not exist. Polyvinylidene fluoride (PVDF) film is a very thin and flexible film that is widely used for film transducer or speaker elements [21]. This piezoelectric polymer is good for applications where mechanical loads are applied [22]. In particular, it can be
0018-9294 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications standards/publications/rights/index.html for more information.
2126
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 7, JULY 2014
applied where signal-to-noise requirements influence very low mass loading by the sensors. In previous studies, a PVDF film was used as a sensor for recoding several biosignals such as respiration [23]–[25], heart rate [26], and BCG [27], [28]. Specifically, in 2007, the reliability and validity of many alternative sensors (including a PVDF sensor) for measuring apnea and hypopnea were reported, along with their scoring approaches, in an American Academy of Sleep Medicine (AASM) paper [29]. Berry et al. compared the capability of a PVDF thermal sensor attached to the upper lip with a pneumotachograph to detect respiratory events in obstructive sleep apnea (OSA) patients [30]. In addition, Koo et al. used PVDF-incorporated belts surrounding the chest and abdomen to validate respiratory event classification during PSG [31]. Although these studies constrainedly measured a subject’s physiological signals to detect sleep apnea, to the best of our knowledge, fully unconstrained systems or methods for monitoring apneic events during sleep using signals measured by PVDF-based sensors have rarely been studied. This study was conducted to establish an unconstrained sleep apnea monitoring method using a PVDF film-based sensor for the continuous and accurate monitoring of apneic events occurring during sleep.
TABLE I SUMMARY OF SUBJECT- AND PSG-RELATED PARAMETERS (MEAN ± STANDARD DEVIATION)
II. MATERIALS AND METHODS A. Participants and PSG Data Thirty-two subjects participated in this study. Twenty-eight nocturnal PSG datasets were collected at the Seoul National University Hospital (SNUH) from July 2012 to February 2013. Four diurnal PSG datasets were additionally recorded for the relatively young ( 0.4 × maximum output voltage (for movement). Condition 2: σm > 0.7 × adaptive threshold (for normal breathing). Condition 3: 0.1 × adaptive threshold < σm < 0.7 × adaptive threshold (for apneic event).
2128
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 7, JULY 2014
Fig. 3. Apneic event decision procedure for every 60-s period. The procedure was based on the following four steps: 1) extraction of the respiratory signal from PVDF data; 2) principal component extraction and data segmentation; 3) threshold determination; and 4) apneic event decision. All values used in the conditions were determined based on the training set data. For example, the subfigure at the lower left corner shows value determination for the upper threshold of the apneic event decision.
Condition 4: σm < 0.1 × adaptive threshold (for out-of-bed event). All of the values used for the conditions were based on the training set data. For instance, the upper threshold was determined by the mean performance evaluation of the training set for an apneic event decision. As shown in the lower left corner of Fig. 3, when the value for the upper threshold of an apneic event decision was 0.7, the mean performance was the highest. Interestingly, this 70% threshold level for an apneic event decision corresponded to standard hypopnea scoring rules [34]. In this analysis, Cohen’s kappa (k) coefficient was marked as the “mean performance.” All of the analyses were performed using MATLAB software (MathWorks Inc., Natick, MA, USA, R2012b version).
D. Statistical Analysis and Evaluation To evaluate the AHI estimation performance, a linear regression analysis was used for the correlation coefficient, and the Bland–Altman method was used to assess the agreement [35]. In this analysis, statistical significance at the 5% level was used. In addition, the sleep apnea diagnosis was conducted based on three levels of AHI cutoff values because there was no threshold value for AHI that clearly discriminated patients with and without sleep apnea [19]. According to the min-by-min analysis, if apneic events occurred more than once (up to six), the current minute was considered to be an “estimated apnea minute (ApneaEST ).” The same rule was applied to the apneic events from PSG (ApneaPSG ), and min-by-min statistical analyses between ApneaEST and ApenaPSG were performed. The sensitivity, specificity, and accuracy were used for these statistical analyses.
In this study, the sensitivity denoted the proportion of correctly identified apnea minutes, while the specificity denoted the proportion of other correctly identified states. In addition, the performance of the proposed method was quantified using Cohen’s kappa, which is very commonly used for studies that measure the agreement between two separate evaluators [36]. III. RESULTS A. Results of AHI Estimation To evaluate the performance of our system, the algorithm determined by the training set data was applied to the test set data, and the apneic event estimation results were compared with the ones from PSG. The AHI from the proposed method (AHIEST ) was compared with the one from PSG (AHIPSG ). As shown in Fig. 4(a), a significant correlation (Pearson’s correlation coefficient = 0.94, p < 0.001) was found between AHIEST and AHIPSG . Fig. 4(b) shows the agreement between AHIEST and AHIPSG , which was evaluated by using the Bland–Altman method. In this figure, the mean difference between the AHI values was −2.3 events/h (no significant difference) and approximately 96% (25 of 26) of the cases were within the dashed lines (95% confidential interval, –15.0–10.2). Table II shows a comparison between the AHI estimation results from the proposed method and those from previous methods. The results from the nasal-oral airflow-based [20] and PVDF thermal sensor-based methods [37] had similar performances compared with ours. In other studies, a nasal airflow-based method [38] showed a higher correlation coefficient than our method, while a peripheral arterial tonometry-based method [19] showed a lower correlation coefficient than the one from ours. In Table II, concordance means the ratio of cases that were within two standard deviations in the Bland–Altman plot.
HWANG et al.: UNCONSTRAINED SLEEP APNEA MONITORING USING POLYVINYLIDENE FLUORIDE FILM-BASED SENSOR
2129
TABLE III DIAGNOSTIC RESULTS AT THREE AHI THRESHOLD LEVELS
operating characteristics curve (ROC) was greater than 0.98. At AHI thresholds of 5 and 15, there was one false positive in each case. However, the estimated AHIs slightly exceeded each AHI cutoff level (5.1 and 15.3). C. Results of Min-By-Min Sleep Apnea Detection
Fig. 4. (a) Relationship between AHIs from PSG and our method. A significant (p < 0.001) correlation was found between these AHIs. (b) Bland–Altman plots between AHIs from PSG and our method. All but one of the cases were distributed within two standard deviations. TABLE II RELATIONSHIP AMONG AHIS ESTIMATED BY EACH METHOD AND PSG
The min-by-min apnea detection results are shown in Table IV. From the test set data, each statistical result was calculated depending on the severity of AHI. From the normal (AHI < 5) to severe level (AHI > 30), the specificity and accuracy were gradually decreased. A kappa statistical analysis revealed a borderline case between substantial (0.6 < k < 0.8) and moderate agreement (0.4 < k < 0.6), whereby the overall k = 0.6. Fig. 5 shows apnea minute estimation results for the best case (see Fig. 5(a), subject #5) and worst case (see Fig. 5(b), subject #21) from the nocturnal data. In the best case, the AHI, sensitivity, specificity, accuracy, and kappa statistic were 46.2 events/h, 91.9%, 88.7%, 90.8%, and 0.79, respectively. In the worst case, the corresponding values were 5.1 events/h, 82.9%, 87.7%, 87.3%, and 0.45, respectively. IV. DISCUSSION A. Agreement Between the Proposed Method and PSG
B. Results of Diagnosing Sleep Apnea With dichotomous AHI thresholds of 5, 15, and 20 events per hour, the sleep apnea diagnosis results for the test set data were assessed using various statistical values. As shown in Table III, for all of the AHI thresholds, the kappa statistic revealed almost perfect agreement (k > 0.8), and the area under the receiver
In this study, we established a fully unconstrained sleep apnea monitoring method using a PVDF film-based sensor. Respiratory signals were obtained from the subjects without their even being aware of the recording process, and apneic events during sleep were automatically detected using the suggested algorithm. The apnea detection process was based on the following steps: 1) the extraction of the respiratory signal from the PVDF data; 2) principal component extraction and data segmentation; 3) threshold determination; and 4) apneic event decisions. When the proposed method was applied to the test set data, it was shown that the estimated AHI from the present study was significantly correlated with the one from PSG [see Fig. 4(a)]. In the min-by-min analysis (see Table IV), for all of the subjects, the kappa statistics revealed greater than moderate agreement (k > 0.4). Furthermore, about half (12 of 26) of the subjects showed substantial agreement (k > 0.6) in the test set. Other results (see Table III) showed that the suggested algorithm
2130
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 7, JULY 2014
TABLE IV STATISTICAL RESULTS OF MIN-BY-MIN SLEEP APNEA DETECTION
B. Comparison With Previous Studies
Fig. 5. (a) Min-by-min apnea detection results for best case (subject #5). The AHI, sensitivity, specificity, accuracy, and kappa statistic values were 46.2 events/h, 91.9%, 88.7%, 90.8%, and 0.79, respectively. (b) Results for worst case (subject #21). The corresponding values were 5.1 events/h, 82.9%, 87.7%, 87.3%, and 0.45, respectively.
could accurately diagnose the sleep apnea patients based on the high accuracy and agreement of the results. Consequently, we can conclude that the results of our method were comparable to those of PSG. Because the PVDF film-based method is simple and does not require trained experts, if combined with signal processing unit, it can be used for ambulatory sleep apnea monitoring. Moreover, it can support apnea event detection during PSG recording.
The proposed unconstrained PVDF-based method had an accuracy similar to that for the ambulatory device that is currently used in clinical practice or constrained PVDF-based methods (see Table II). For instance, Ayas et al. used a wrist-worn device that combines a peripheral arterial tonometer (PAT), actigraph, and arterial oxygen saturation to diagnose OSA [19]. Even though the device used the attenuation of the PAT signal amplitude, which is strongly correlated with apnea [39], along with the SaO2 signal, which directly reflects the absence of breathing, the correlation coefficient between PSG AHI and the wrist-worn device was lower than that for ours. Another example is the NightWatch (NW) system, which records the SaO2 , nasal-oral airflow, and chest and abdominal wall motion using sensors attached to the patient’s face and body [20]. White et al. assessed the accuracy of the NW system used at home and in the lab to monitor sleep apnea. Although NW collected many physiological signals for apnea detection, its AHI estimation performance was similar to our results. However, in the present study, the method and system were not applied at home, and the apnea detection accuracy might be reduced under this condition. Han et al. also detected apneic events using a single nasal airflow channel from PSG data [38]. The performance of their method was found to be better than ours in a linear regression analysis of AHI, while the concordance on the Bland–Altman plot calculated in this study was slightly higher than that found in their study. Koo et al. detected respiratory events using PVDF impedance belts surrounding the patient’s chest and abdomen [31], and their method was comparable to standard respiratory inductance plethysmography in determining respiratory events during PSG. Nakano et al. monitored oronasal airflow using a PVDF thermal sensor, and found that an airflow monitor could be used to detect SDB [37]. Berry et al. compared the readings from a PVDF thermal sensor attached to the upper lip with a mask pneumotachograph and accurately detected respiratory events compared with the detection accuracy of pneumotachography in patients with OSA [30]. Even though they used belts or a thermal sensor that could directly measure the respiratory-induced signals, there were no significant differences in the AHI estimation results between the constrained methods and our unconstrained method. In particular, the apnea diagnostic ability at a fixed AHI threshold of the proposed method performed better in relation to the sensitivity, specificity, kappa statistics, and area
HWANG et al.: UNCONSTRAINED SLEEP APNEA MONITORING USING POLYVINYLIDENE FLUORIDE FILM-BASED SENSOR
under the ROC than the other methods with a PVDF-based sensor [31], [37]. Moreover, the greatest disadvantage of these systems was the necessity for the subject to wear or attach the PVDF-based sensor to their body or face during sleep, which could interrupt their normal sleep architecture. However, the number of analyzed apnea datasets in this study was less than those of the previous PVDF-based studies [31], [37], and our apnea detection performance could be different with larger data sets. C. Validation of PVDF Film-Based Sensors In this study, the PVDF sensors were uniformly oriented under the subject’s back position in a particular direction. To ensure the validity and reliability of the respiratory signal measurements, the PVDF sensors were aligned horizontally, as shown in Fig. 1, because subjects tend to move more often from side to side than in the transverse planar direction (up and down) during sleep. Sensors that are similar in shape and size to ours can easily be found in several commercialized sleep monitoring devices. For example, SleepScan SL-501 (TANITA, Tokyo, Japan) is a device that collects numerous kinds of physiological signals during sleep, including the pulse rate, respiration rate, and body motion data, and it is designed to be long in the horizontal direction. Other analogous examples are the EarlySense System (EarlySense, Waltham, MA, USA) and Nemuri (which means “Sleep”) SCAN NN-1300 (PARAMOUNT BED, Tokyo, Japan). Using PVDF sensors, we could obtain not only respiration signals, but also BCG signals. Moreover, when the PVDF sensor was aligned horizontally, the best signal-to-noise ratio (SNR) for the BCG signal was revealed in the preliminary test. It is speculated that blood ejection-induced vibration was transferred more strongly along a particular axis to the sensor through the bed mattress during the recording. However, the results of the SNR test that depended on the PVDF direction were not included in this paper. As mentioned in Section II, the data from PVDF channels #1 and #2 that reflected changes in the chest volume induced by breathing were not selected for the final analysis. From the training set data, the apnea detection performances using all of the channels were lower than those from channels #3 and #4. Because the subjects slept most of the time with pillows during PSG, there was a small gap between the bed mattress and a subject’s upper body. One could speculate that the volume change in the chest during respiration did not effectively transfer to the PVDF because the upper body was not fully in contact with the sensor. Furthermore, a PVDF impedance belt-based study showed that the respiratory signal measured by a chest belt was less correlated with the airflow than that from an abdomen belt [31]. As a result, channels #1 and #2 showed relatively high error components in the PVDF signals, which could degrade the apnea detection performance. This is why the data from these channels were finally excluded in the analysis. D. Validation of Apnea Detection Algorithm During PSG recording, the respiration patterns were different for the various subjects and varied depending on the position
2131
in bed and the sleep posture. To consider these things, adaptive thresholds were set based on the standard deviation of the PVDF signals for every minute for each subject in the apnea detection algorithm. As a result, we could obtain accurate AHI estimation results compared with PSG AHI. However, our method tended to underestimate apneic events at a higher AHI level, as shown in Fig. 4(a) and (b). At this level, apneic events occurred frequently in a short period, and the apneic event detection was ineffective under these conditions. In this study, the apnea detection algorithm was developed based on a “fixed time window” in consideration of real-time processing for ambulatory or home monitoring purposes. Apnea can persist for more than 30 s (1 epoch) in patients with severe apnea, and these durations can be diagnosed as normal breathing in our algorithm because the apnea threshold was set based on a change in the standard deviation of the PVDF data in the fixed time window. This is why we used a 60-s analysis window instead of 1 epoch, which is the most widely used time scale in sleep studies. Despite these efforts, the increased baseline of the threshold due to the restricted time window made it difficult to detect consecutive multiple apneic events. In contrast, our method tended to overestimate apneic events at the lower AHI level. One could speculate that our method may misestimate a respiratory signal drop as an apneic event that does not meet the standard scoring criteria. Moreover, other SRBD such as snoring, respiratory event-related arousals, upper airway obstruction, and Cheyne–Stokes breathing could occur in patients with sleep apnea [2]. These can also influence the amplitude decrease in the respiratory signals and might be estimated to be sleep apnea or hypopnea in our system. Moreover, PVDF signals may be distorted by motion artifacts that cannot be removed completely from the analysis process, which is one of the reasons that our method overestimates apneic events at a lower AHI. Despite these shortcomings, the overall apnea detection performance showed a high relationship between AHIs (R = 0.94). In addition, as shown in Table IV, the sensitivity was significantly lower than the specificity in the normal and mild severity groups. In these groups, the apnea detection sensitivity could be reduced considerably by only a few wrong estimates because the percentage of sleep apnea occurrences was markedly less than that of normal breathing. Thus, the kappa statistic was also used to evaluate the apnea detection performance. In particular, the difference between the sensitivity and specificity of the normal group was approximately 10% less than that of the moderate severity group, but there was no difference between the overall kappa statistics. Therefore, we could conclude that the proposed method showed similar apnea detection performances in all of the groups. In this study, the algorithm performed poorly when subjects slept in a lateral posture. During the respiratory cycle, respiratory-induced vertical (frontal axis) pressure was effectively transferred from the body to the PVDF sensors in a supine or prone posture. However, respiratory-related vertical pressure was transferred horizontally to the PVDF sensor in the lateral posture, and the respiratory signal measured by the sensor was significantly attenuated or distorted. As a result, the apnea detection performance in the lateral posture was
2132
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 7, JULY 2014
not validated for subjects who shared the bed with the partner. Additional PVDF sensor or increased size of sensor will probably be needed to detect sleep apnea for two people. Lastly, the practicability of the proposed method was not assessed in the home and ambulatory environments. In these environments, the apnea detection performance could be different, and problems that were not considered in the sleep laboratory might occur. To verify this issue, a validation of the proposed algorithm will be performed using PVDF signals measured in the home and ambulatory environments in a future study. V. CONCLUSION Fig. 6. PC1R P V D F , PC1R E S P , respiration from thoracic movement, respiration from nasal-oral airflow and apnea events, respectively. Amplitude of PC1R E S P is decreased with oscillation during OSA. On the other hand, amplitude of PC1R E S P is also decreased but without oscillation during CSA. In this figure, data from subject #29 were used.
relatively low and reflected a tradeoff between detection accuracy and unconstrained monitoring. In this study, very minute levels of oscillations such as “respiratory efforts” that can separate the OSA and central sleep apnea (CSA) could be measured using PVDF sensor, as shown in Fig. 6. In the figure, amplitude of respiration signal from nasal airflow sensor is decreased and that from chest belt sensor is decreased with oscillation during OSA. On the other hand, the amplitude from nasal airflow sensor has similar trend to OSA, but the amplitude of signal from chest belt sensor is decreased without oscillation during CSA. Those differences between OSA and CSA are also shown through our PVDF film-based system. So, we may think that OSA and CSA can be distinguished in our system using this amplitude decimation attenuation of the signal amplitude and oscillation occurrence. Although types of sleep apnea were not separated in this study, we plan to focus on OSA/CSA classification using PVDF film-based system to improve the impact of our work. E. Limitations Our proposed method also had some limitations. First, we did not separate sleep apnea into obstructive, central, and mixed types as mentioned in “Validation of apnea detection algorithm” session. With an improved algorithm and PVDF sensor, OSA and CSA can be divided into different types of apnea. Second, hypopnea events were not differentiated in the apneic events decision. However, this seems to have little clinical significance because both types of events have similar consequences and pathophysiologies [13]. In addition, AHI, which is an index of sleep apnea severity, is defined as the “sum” of apnea and hypopnea events instead of the frequency of each type of event [19]. Third, no analysis of the event-by-event apnea detection compared with PSG was conducted in this study because the coupled systems had different time scales for diagnosing sleep apnea. Fourth, test of the size and shape optimization of PVDF-based sensor was not performed because we were focusing on the possibility and potential of sleep apnea monitoring using PVDF film-based sensor in this study. Fifth, the proposed method was
In this study, a new sleep apnea monitoring method using a PVDF film-based sensor was developed that can unconstrainedly measure respiratory signals from subjects. Using the proposed system and method, we were able to obtain apnea detection performances similar to those of ambulatory devices and constrained PVDF-based studies. The developed system and method can be applied to sleep apnea detection for home or ambulatory monitoring purposes. REFERENCES [1] C. A. Kushida, M. R. Littner, T. Morgenthaler, C. A. Alessi, D. Bailey, J. Coleman, L. Friedman, M. Hirshkowitz, S. Kapen, M. Kramer, T. L. Chiong, D. L. Loube, J. Owens, J. P. Pancer, and M. Wise, “Practice parameters for the indications for polysomnography and related procedures: An update for 2005,” Sleep, vol. 28, no. 4, pp. 499–521, 2005. [2] M. H. Kryger, T. Roth, and W. C. Dement, “Normal human sleep: An overview,” in Principles and Practice of Sleep Medicine, 5th ed. Philadelphia, PA, USA: Saunders, 2011, pp. 16–26. [3] R. L. Wilkin, Clinical Assessment in Respiratory Care. St. Louis, MO, USA: Mosby, 1985, pp. 268–272. [4] B. El-Ad and P. Lavie, “Effect of sleep apnea on cognition and mood,” Int. Rev. Psychiatry, vol. 17, no. 4, pp. 277–282, 2005. [5] M. Hirshkowitz, I. Karacan, A. Gurakar, and R. L. Williams, “Hypertension, erectile dysfunction, and occult sleep apnea,” Sleep, vol. 12, no. 3, pp. 223–232, 1989. [6] S. Yan-fang and W. Yu-ping, “Sleep-disordered breathing: Impact on functional outcome of ischemic stroke patients,” Sleep Med., vol. 10, no. 7, pp. 717–719, 2009. [7] R. S. Leung, “Sleep-disordered breathing: Autonomic mechanisms and arrhythmias,” Prog. Cardiovasc. Dis., vol. 51, no. 4, pp. 324–338, 2009. [8] H. E. Resnick, S. Redline, E. Shahar, A. Gilpin, A. Newman, R. Walter, G. A. Ewy, B. V. Howard, and N. M. Punjabi, “Diabetes and sleep disturbances: Findings from the sleep heart health study,” Diabetes Care, vol. 26, no. 3, pp. 702–709, 2003. [9] Y. Peker, J. Hedner, A. Johansson, and M. Bende, “Reduced hospitalization with cardiovascular and pulmonary disease in obstructive sleep apnea patients on nasal CPAP treatment,” Sleep, vol. 20, no. 8, pp. 645–653, 1997. [10] T. Young, M. Palta, J. Dempsey, J. Skatrud, S. Weber, and S. Badr, “The occurrence of sleep-disordered breathing among middle-aged adults,” New Engl. J. Med., vol. 328, no. 17, pp. 1230–1235, 1993. [11] J. K. Kim, K. H. In, J. H. Kim, S. H. You, K. H. Kang, J. J. Shim, S. Y. Lee, J. B. Lee, S. G. Lee, C. Park, and C. Shin, “Prevalence of sleep-disordered breathing in middle-aged korean men and women,” Amer. J. Respir. Crit. Care. Med., vol. 170, no. 10, pp. 1108–1113, 2004. [12] P. E. Peppard, T. Young, J. H. Barnet, M. Palta, E. W. Hagen, and K. M. Hla, “Increased prevalence of sleep-disordered breathing in adults,” Amer. J. Epidemiol., vol. 177, no. 9, pp. 1006–1014, 2013. [13] “Sleep-related breathing disorders in adults: Recommendations for syndrome definition and measurement techniques in clinical research. The report of an American academy of sleep medicine task force,” Sleep, vol. 22, no. 5, pp. 667–689, 1999. [14] J. Boyle, N. Bidargaddi, A. Sarela, and M. Karunanithi, “Automatic detection of respiration rate from ambulatory single-lead ECG,” IEEE Trans. Inf. Technol. Biomed., vol. 13, no. 6, pp. 890–896, Nov. 2009.
HWANG et al.: UNCONSTRAINED SLEEP APNEA MONITORING USING POLYVINYLIDENE FLUORIDE FILM-BASED SENSOR
[15] M. O. Mendez, A. M. Bianchi, M. Matteucci, S. Cerutti, and T. Penzel, “Sleep apnea screening by autoregressive models from a single ECG lead,” IEEE Trans. Biomed. Eng., vol. 56, no. 12, pp. 2838–2850, Dec. 2009. [16] P. de Chazal, C. Heneghan, E. Sheridan, R. Reilly, P. Nolan, and M. O’Malley, “Automated processing of the single-lead electrocardiogram for the detection of obstructive sleep apnoea,” IEEE Trans. Biomed. Eng., vol. 50, no. 6, pp. 686–696, Jun. 2003. [17] F. Roche, J. M. Gaspoz, I. C. Fortune, P. Minini, V. Pichot, D. Duverney, F. Costes, J. R. Lacour, and J. C. Barth´el´emy, “Screening of obstructive sleep apnea syndrome by heart rate variability analysis,” Circulation, vol. 100, no. 13, pp. 1411–1415, 1999. [18] Y. G. Lim, K. H. Hong, K. K. Kim, J. H. Shin, S. M. Lee, G. S. Chung, H. J. Baek, D. U. Jeong, and K. S. Park, “Monitoring physiological signals using nonintrusive sensors installed in daily life equipment,” Biomed. Eng. Lett., vol. 1, pp. 11–20, 2011. [19] N. T. Ayas, S. Pittman, M. MacDonald, and D. P. White, “Assessment of a wrist-worn device in the detection of obstructive sleep apnea,” Sleep Med., vol. 4, no. 5, pp. 435–442, 2003. [20] D. P. White, T. J. Gibb, J. M. Wall, and P. R. Westbrook, “Assessment of accuracy and analysis time of a novel device to monitor sleep and breathing in the home,” Sleep, vol. 18, no. 2, pp. 115–126, 1995. [21] M. A. O’Reilly and K. Hynynen, “A PVDF receiver for ultrasound monitoring of transcranial focused ultrasound therapy,” IEEE Trans. Biomed. Eng., vol. 57, no. 9, pp. 2286–2294, Sep. 2010. [22] S. H. Hwang, H. N. Yoon, D. W. Jung, Y. J. Lee, D. U. Jeong, and K. S. Park, “Apnea event estimation during sleep using polyvinylidene fluoride film,” in Proc. eTELEMED, 2013, pp. 280–283. [23] J. Siivola, “New noninvasive piezoelectric transducer for recording of respiration, heart rate and body movements,” Med. Biol. Eng. Comput., vol. 27, no. 4, pp. 423–425, 1989. [24] G. R. Manjunatha, K. Rajanna, D. R. Mahapatra, M. M. Nayak, U. M. Krishnaswamy, and R. Srinivasa, “Polyvinylidene fluoride film based nasal sensor to monitor human respiration pattern: An initial clinical study,” J. Clin. Monit. Comput., vol. 27, pp. 647–657, Jun. 2013. [25] S. J. Choi and Z. W. Jiang, “A novel wearable sensor device with conductive fabric and PVDF film for monitoring cardiorespiratory signals,” Sens. Actuators, A, vol. 128, no. 2, pp. 317–326, 2006. [26] F. Wang, M. Tanaka, and S. Chonan, “Development of a PVDF piezopolymer sensor for unconstrained in-sleep cardiorespiratory monitoring,” J. Intel. Mater. Syst. Struct., vol. 14, no. 3, pp. 185–190, 2003. [27] H. J. Baek, G. S. Chung, K. K. Kim, and K. S. Park, “A smart health monitoring chair for nonintrusive measurement of biological signals,” IEEE Trans. Inf. Technol. Biomed., vol. 16, no. 1, pp. 150–158, Jan. 2012. [28] Y. Y. Chiu, W. Y. Lin, H. Y. Wang, S. B. Huang, and M. H. Wu, “Development of a piezoelectric polyvinylidene fluoride (PVDF) polymer-based sensor patch for simultaneous heartbeat and respiration monitoring,” Sens. Actuators A, vol. 189, pp. 328–334, 2013. [29] S. Redline, R. Budhiraja, V. Kapur, C. L. Marcus, J. H. Mateika, R. Mehra, S. Parthasarthy, V. K. Somers, K. P. Strohl, L. G. Sulit, D. Gozal, M. S. Wise, and S. F. Quan, “The scoring of respiratory events in sleep: Reliability and validity,” J. Clin. Sleep. Med., vol. 3, no. 2, pp. 169–200, 2007. [30] R. B. Berry, G. L. Koch, S. Trautz, and M. H. Wagner, “Comparison of respiratory event detection by a polyvinylidene fluoride film airflow sensor and a pneumotachograph in sleep apnea patients,” Chest, vol. 128, no. 3, pp. 1331–1338, 2005. [31] B. B. Koo, C. Drummond, S. Surovec, N. Johnson, S. A. Marvin, and S. Redline, “Validation of a polyvinylidene fluoride impedance sensor for respiratory event classification during polysomnography,” J. Clin. Sleep. Med., vol. 7, no. 5, pp. 479–485, 2011. [32] A. Rechtschaffen and A. Kales, A Manual of Standardized Terminology, Techniques and Scoring System for Sleep Stages of Human Subjects. Washington, DC, USA: NIH Publication, 1968. [33] R. O. Duda, P. E. Hart, and D. G. Stork, Pattern Classification, 2nd ed. Hoboken, NJ, USA: Wiley, 2001. [34] C. Iber, The AASM Manual for the Scoring of Sleep and Associated Events: Rules, Terminology and Technical Specifications. Westchester, IL, USA: Amer. Acad. Sleep Med., 2007, pp. 45–47. [35] J. M. Bland and D. G. Altman, “Statistical methods for assessing agreement between two methods of clinical measurement,” Lancet, vol. 1, no. 8476, pp. 307–310, 1986. [36] A. J. Viera and J. M. Garrett, “Understanding interobserver agreement: The kappa statistic,” Family Med., vol. 37, no. 5, pp. 360–363, 2005. [37] H. Nakano, T. Tanigawa, T. Furukawa, and S. Nishima, “Automatic detection of sleep-disordered breathing from a single-channel airflow record,” Eur. Respir. J., vol. 29, no. 4, pp. 728–736, 2007.
2133
[38] J. H. Han, H. B. Shin, D. U. Jeong, and K. S. Park, “Detection of apneic events from single channel nasal airflow using 2nd derivative method,” Comput. Meth. Programs Biomed., vol. 91, no. 3, pp. 199–207, 2008. [39] R. P. Schnall, A. Shlitner, J. Sheffy, R. Kedar, and P. Lavie, “Periodic, profound peripheral vasoconstriction—A new marker of obstructive sleep apnea,” Sleep, vol. 22, no. 7, pp. 939–946, 1999.
Su Hwan Hwang received the B.S. degree in biomechatronics and electronic electrical engineering from Sungkyunkwan University, Seoul, Korea, in 2009. He is currently working toward the combined M.Sc. and Ph.D. degrees in interdisciplinary program in bioengineering from Seoul National University, Seoul, Korea. His current research interests include biomedical signal measurement and processing in sleep medicine with the goal of translating the engineering works into practical medical research and clinical applications.
Hong Ji Lee received the B.S. degree in information communication engineering from Ewha Womans University, Seoul, Korea, in 2010. She is currently working toward the combined M.Sc. and Ph.D. degrees in interdisciplinary program in bioengineering from Seoul National University, Seoul, Korea. Her main research interests include the study of biomedical instrumentations and signal processing for ubiquitous healthcare. Her current research interests include tremor monitoring systems for diagnosis of Parkinson disease.
Hee Nam Yoon received the B.S. degree in biomedical engineering from Kyung Hee University, Yongin, Korea, in 2011. He is currently working toward the combined M.Sc. and Ph.D. degrees in interdisciplinary program in bioengineering from Seoul National University, Seoul, Korea. His current research interests include biomedical signals and systems for sleep medicine and synchronization dynamics of human biological signals.
Da Woon Jung received the B.S. degree in biomedical engineering from Kyung Hee University, Yongin, Korea, in 2011. She is currently working toward the combined M.Sc. and Ph.D. degrees in interdisciplinary program in bioengineering from Seoul National University, Seoul, Korea. Her current research interests include biomedical signals analysis for sleep medicine and synchronization dynamics of human biological signals.
2134
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 7, JULY 2014
Yu-Jin G. Lee received the M.D. degree from Soonchunhyang University Hospital, Cheonan, Korea, in 2005. She is a Psychiatrist and a Specialist in sleep medicine. She was a Clinical Fellow at the Center for Sleep and Chronobiology, Seoul National University Hospital. She is currently with the Seoul Metropolitan Eunpyeong Hospital, Seoul, Korea. Dr. Lee is a Member of the Board of Directors of the Korean Academy of Sleep Medicine.
Yu Jin Lee received the M.D. degree from Ewha Womans University, Seoul, Korea and the Ph.D. degree from the Chungbuk National University, Cheongju, Korea. She is currently an Associate Professor in the Division of Sleep and Chronobiology, Department of Psychiatry, College of Medicine and Hospital, Seoul National University Seoul, Korea.
Do-Un Jeong received the M.D. and Ph.D. degrees from Seoul National University, Seoul, Korea, in 1976 and 1988, respectively. He is currently a Professor of psychiatry at the Seoul National University College of Medicine and Director of the Center for Sleep and Chronobiology, Seoul National University Hospital. His research interests include signal processing in sleep and clinical sleep disorders. Dr. Jeong is the former President of the Korean Society of Medical and Biological Engineering and the Korean Academy of Sleep Medicine as well as a Fellow of the American Academy of Sleep Medicine.
Kwang Suk Park (M’78–SM’09) received the B.S., M.S., and Ph.D. degrees in electronics engineering from Seoul National University, Seoul, Korea, in 1980, 1982, and 1985, respectively. In 1985, he joined the Department of Biomedical Engineering, College of Medicine, Seoul National University, Seoul, as a Founding Staff Member, where he is currently a Professor. He is also the Director of Advanced Biometric Research Center, Seoul National University. His main research interests include biological signal measurement and processing for the diagnosis. His current research interests include nonintrusive measurements of biological signals for ubiquitous healthcare. Dr. Park is a member of the Korean Society of Medical and Biological Engineering and has served as the Secretary General of the World Congress on Medical Physics and Biomedical Engineering held in Seoul in 2006. He is also a Senior Member of the IEEE Engineering in Medicine and Biology Society. Since 2005, has been serving as an Associated Editor for the IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS. He has also chaired or co-chaired the Annual International Conference on u-Healthcare during past ten years.
2125
Unconstrained Sleep Apnea Monitoring Using Polyvinylidene Fluoride Film-Based Sensor Su Hwan Hwang, Hong Ji Lee, Hee Nam Yoon, Da Woon Jung, Yu-Jin G. Lee, Yu Jin Lee, Do-Un Jeong, and Kwang Suk Park∗ , Senior Member, IEEE
Abstract—We established and tested an unconstrained sleep apnea monitoring method using a polyvinylidene (PVDF) filmbased sensor for continuous and accurate monitoring of apneic events occurred during sleep. Twenty-six sleep apnea patients and six normal subjects participated in this study. Subjects’ respiratory signals were measured using the PVDF-based sensor during polysomnography. The PVDF sensor comprised a 4 × 1 array, and a thin silicon pad was placed over the sensor to prevent damage. Total thickness of the merged system was approximately 1.1 mm which was thin enough to prevent the subject from being consciously aware of its presence. It was designed to be placed under subjects’ backs and installed between a bed cover and mattress. The proposed method was based on the standard deviation of the PVDF signals, and it was applied to a test set for detecting apneic events. The method’s performance was assessed by comparing the results with a sleep physician’s manual scoring. The correlation coefficient for the apnea-hypopnea index (AHI) values between the methods was 0.94 (p < 0.001). The areas under the receiver operating curves at three AHI threshold levels (>5, >15, and >20) for sleep apnea diagnosis were 0.98, 0.99, and 0.98, respectively. For min-by-min apnea detection, the method classified sleep apnea with an average sensitivity of 72.9%, specificity of 90.6%, accuracy of 85.5%, and kappa statistic of 0.60. The developed system and method can be applied to sleep apnea detection in home or ambulatory monitoring. Index Terms—Apnea-hypopnea index (AHI), PVDF-sensor, unconstrained sleep apnea monitoring.
I. INTRODUCTION LEEP APNEA is a typical sleep-related breathing disorder (SRBD) characterized by frequent, abnormal cessation of respiration during sleep [1]. During the apneic period, there is an increased effort in breathing, leading to arousal and sleep fragmentation [2]. Thus, severe and frequent sleep apnea disrupts
S
Manuscript received July 9, 2013; revised November 4, 2013; accepted March 25, 2014. Date of publication April 4, 2014; date of current version June 14, 2014. This work was supported by Samsung Electronics, Inc. Asterisk indicates corresponding author. S. H. Hwang, H. J. Lee, H. N. Yoon, and D. W. Jung are with the Interdisciplinary Program in Bioengineering, Seoul National University, Seoul 110-799, Korea (e-mail: [email protected]; [email protected]; [email protected]; [email protected]). Y.-J. G. Lee is with the Seoul Metropolitan Eunpyeong Hospital, Seoul 122-913, Korea (e-mail: [email protected]). Y. J. Lee and D.-U. Jeong are with the Department of Neuropsychiatry and the Center for Sleep and Chronobiology, Seoul National University Hospital, Seoul 110-744, Korea (e-mail: [email protected]; [email protected]). ∗ K. S. Park is with the Department of Biomedical Engineering, College of Medicine, Seoul National University, Seoul 110-799, Korea (e-mail: [email protected]). Digital Object Identifier 10.1109/TBME.2014.2314452
the sleep architecture of subjects and can lead to sleep disorders such as severe snoring [3], fatigue, daytime sleepiness [4], and systemic hypertension [5]. In addition, apnea-induced hypoxia during sleep can cause stroke [6], arrhythmias [7], diabetes [8], and cardiovascular diseases [9]. In previous study, Young et al. reported that approximately 2% of adult women and 4% of adult men in the middle-age range are affected by sleep apnea [10]. Kim et al. analyzed data to determine the prevalence of sleep-disordered breathing (SDB) in 457 Korean adults aged 40–69 years and 27% of men and 16% of women, respectively, had SDB [11]. Furthermore, a recent study of SDB in adults discovered the actual prevalence rates of SDB representing substantial increase over the last two decades (10% among 30–49-year-old men; 17% among 50–70-year-old men; 3% among 30–49-year-old women; and 9% among 50–70-yearold women) [12]. In a general sleep and sleep-related disorder monitoring system, polysomnography (PSG) has been regarded as the goldstandard method. According to the PSG recording and diagnosis rules, apneic events are detected based on the nasal-oral airflow amplitude and blood oxygen saturation (SaO2 ) level [13]. Even though PSG has been used for the assessment of sleep apnea, the detection of apneic events using PSG has some problems. First, PSG recording during sleep is an inconvenient experience for subjects because numerous sensors are attached to the patient’s body (SpO2 ) and face (nasal-oral airflow). Second, the manual scoring of nocturnal apneic events from PSG data is a very time-consuming and laborious process, and it demands specially trained sleep experts. For overcoming some of these drawbacks, in recent years, many alternative methods without using PSG to detect apneic or hypopneic events have been proposed. These studies have involved sleep apnea detection based on several biosignals. For example, electrocardiogram (ECG)-based studies have shown that RR-interval- or R-peak amplitude-based methods [14]–[17] are useful for apnea detection. A ballistocardiogram (BCG)-based system measured the respiration rate with high correlation using an air mattress with a balancing tube [18]. Pulse oximetry- and respiratory-based studies [19], [20] revealed a relatively high correlation coefficient (r > 0.9) between the apnea-hypopnea index (AHI) from PSG and the suggested one. Despite these efforts, a clearly superior method or system for unconstrained sleep apnea monitoring still does not exist. Polyvinylidene fluoride (PVDF) film is a very thin and flexible film that is widely used for film transducer or speaker elements [21]. This piezoelectric polymer is good for applications where mechanical loads are applied [22]. In particular, it can be
0018-9294 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications standards/publications/rights/index.html for more information.
2126
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 7, JULY 2014
applied where signal-to-noise requirements influence very low mass loading by the sensors. In previous studies, a PVDF film was used as a sensor for recoding several biosignals such as respiration [23]–[25], heart rate [26], and BCG [27], [28]. Specifically, in 2007, the reliability and validity of many alternative sensors (including a PVDF sensor) for measuring apnea and hypopnea were reported, along with their scoring approaches, in an American Academy of Sleep Medicine (AASM) paper [29]. Berry et al. compared the capability of a PVDF thermal sensor attached to the upper lip with a pneumotachograph to detect respiratory events in obstructive sleep apnea (OSA) patients [30]. In addition, Koo et al. used PVDF-incorporated belts surrounding the chest and abdomen to validate respiratory event classification during PSG [31]. Although these studies constrainedly measured a subject’s physiological signals to detect sleep apnea, to the best of our knowledge, fully unconstrained systems or methods for monitoring apneic events during sleep using signals measured by PVDF-based sensors have rarely been studied. This study was conducted to establish an unconstrained sleep apnea monitoring method using a PVDF film-based sensor for the continuous and accurate monitoring of apneic events occurring during sleep.
TABLE I SUMMARY OF SUBJECT- AND PSG-RELATED PARAMETERS (MEAN ± STANDARD DEVIATION)
II. MATERIALS AND METHODS A. Participants and PSG Data Thirty-two subjects participated in this study. Twenty-eight nocturnal PSG datasets were collected at the Seoul National University Hospital (SNUH) from July 2012 to February 2013. Four diurnal PSG datasets were additionally recorded for the relatively young ( 0.4 × maximum output voltage (for movement). Condition 2: σm > 0.7 × adaptive threshold (for normal breathing). Condition 3: 0.1 × adaptive threshold < σm < 0.7 × adaptive threshold (for apneic event).
2128
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 7, JULY 2014
Fig. 3. Apneic event decision procedure for every 60-s period. The procedure was based on the following four steps: 1) extraction of the respiratory signal from PVDF data; 2) principal component extraction and data segmentation; 3) threshold determination; and 4) apneic event decision. All values used in the conditions were determined based on the training set data. For example, the subfigure at the lower left corner shows value determination for the upper threshold of the apneic event decision.
Condition 4: σm < 0.1 × adaptive threshold (for out-of-bed event). All of the values used for the conditions were based on the training set data. For instance, the upper threshold was determined by the mean performance evaluation of the training set for an apneic event decision. As shown in the lower left corner of Fig. 3, when the value for the upper threshold of an apneic event decision was 0.7, the mean performance was the highest. Interestingly, this 70% threshold level for an apneic event decision corresponded to standard hypopnea scoring rules [34]. In this analysis, Cohen’s kappa (k) coefficient was marked as the “mean performance.” All of the analyses were performed using MATLAB software (MathWorks Inc., Natick, MA, USA, R2012b version).
D. Statistical Analysis and Evaluation To evaluate the AHI estimation performance, a linear regression analysis was used for the correlation coefficient, and the Bland–Altman method was used to assess the agreement [35]. In this analysis, statistical significance at the 5% level was used. In addition, the sleep apnea diagnosis was conducted based on three levels of AHI cutoff values because there was no threshold value for AHI that clearly discriminated patients with and without sleep apnea [19]. According to the min-by-min analysis, if apneic events occurred more than once (up to six), the current minute was considered to be an “estimated apnea minute (ApneaEST ).” The same rule was applied to the apneic events from PSG (ApneaPSG ), and min-by-min statistical analyses between ApneaEST and ApenaPSG were performed. The sensitivity, specificity, and accuracy were used for these statistical analyses.
In this study, the sensitivity denoted the proportion of correctly identified apnea minutes, while the specificity denoted the proportion of other correctly identified states. In addition, the performance of the proposed method was quantified using Cohen’s kappa, which is very commonly used for studies that measure the agreement between two separate evaluators [36]. III. RESULTS A. Results of AHI Estimation To evaluate the performance of our system, the algorithm determined by the training set data was applied to the test set data, and the apneic event estimation results were compared with the ones from PSG. The AHI from the proposed method (AHIEST ) was compared with the one from PSG (AHIPSG ). As shown in Fig. 4(a), a significant correlation (Pearson’s correlation coefficient = 0.94, p < 0.001) was found between AHIEST and AHIPSG . Fig. 4(b) shows the agreement between AHIEST and AHIPSG , which was evaluated by using the Bland–Altman method. In this figure, the mean difference between the AHI values was −2.3 events/h (no significant difference) and approximately 96% (25 of 26) of the cases were within the dashed lines (95% confidential interval, –15.0–10.2). Table II shows a comparison between the AHI estimation results from the proposed method and those from previous methods. The results from the nasal-oral airflow-based [20] and PVDF thermal sensor-based methods [37] had similar performances compared with ours. In other studies, a nasal airflow-based method [38] showed a higher correlation coefficient than our method, while a peripheral arterial tonometry-based method [19] showed a lower correlation coefficient than the one from ours. In Table II, concordance means the ratio of cases that were within two standard deviations in the Bland–Altman plot.
HWANG et al.: UNCONSTRAINED SLEEP APNEA MONITORING USING POLYVINYLIDENE FLUORIDE FILM-BASED SENSOR
2129
TABLE III DIAGNOSTIC RESULTS AT THREE AHI THRESHOLD LEVELS
operating characteristics curve (ROC) was greater than 0.98. At AHI thresholds of 5 and 15, there was one false positive in each case. However, the estimated AHIs slightly exceeded each AHI cutoff level (5.1 and 15.3). C. Results of Min-By-Min Sleep Apnea Detection
Fig. 4. (a) Relationship between AHIs from PSG and our method. A significant (p < 0.001) correlation was found between these AHIs. (b) Bland–Altman plots between AHIs from PSG and our method. All but one of the cases were distributed within two standard deviations. TABLE II RELATIONSHIP AMONG AHIS ESTIMATED BY EACH METHOD AND PSG
The min-by-min apnea detection results are shown in Table IV. From the test set data, each statistical result was calculated depending on the severity of AHI. From the normal (AHI < 5) to severe level (AHI > 30), the specificity and accuracy were gradually decreased. A kappa statistical analysis revealed a borderline case between substantial (0.6 < k < 0.8) and moderate agreement (0.4 < k < 0.6), whereby the overall k = 0.6. Fig. 5 shows apnea minute estimation results for the best case (see Fig. 5(a), subject #5) and worst case (see Fig. 5(b), subject #21) from the nocturnal data. In the best case, the AHI, sensitivity, specificity, accuracy, and kappa statistic were 46.2 events/h, 91.9%, 88.7%, 90.8%, and 0.79, respectively. In the worst case, the corresponding values were 5.1 events/h, 82.9%, 87.7%, 87.3%, and 0.45, respectively. IV. DISCUSSION A. Agreement Between the Proposed Method and PSG
B. Results of Diagnosing Sleep Apnea With dichotomous AHI thresholds of 5, 15, and 20 events per hour, the sleep apnea diagnosis results for the test set data were assessed using various statistical values. As shown in Table III, for all of the AHI thresholds, the kappa statistic revealed almost perfect agreement (k > 0.8), and the area under the receiver
In this study, we established a fully unconstrained sleep apnea monitoring method using a PVDF film-based sensor. Respiratory signals were obtained from the subjects without their even being aware of the recording process, and apneic events during sleep were automatically detected using the suggested algorithm. The apnea detection process was based on the following steps: 1) the extraction of the respiratory signal from the PVDF data; 2) principal component extraction and data segmentation; 3) threshold determination; and 4) apneic event decisions. When the proposed method was applied to the test set data, it was shown that the estimated AHI from the present study was significantly correlated with the one from PSG [see Fig. 4(a)]. In the min-by-min analysis (see Table IV), for all of the subjects, the kappa statistics revealed greater than moderate agreement (k > 0.4). Furthermore, about half (12 of 26) of the subjects showed substantial agreement (k > 0.6) in the test set. Other results (see Table III) showed that the suggested algorithm
2130
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 7, JULY 2014
TABLE IV STATISTICAL RESULTS OF MIN-BY-MIN SLEEP APNEA DETECTION
B. Comparison With Previous Studies
Fig. 5. (a) Min-by-min apnea detection results for best case (subject #5). The AHI, sensitivity, specificity, accuracy, and kappa statistic values were 46.2 events/h, 91.9%, 88.7%, 90.8%, and 0.79, respectively. (b) Results for worst case (subject #21). The corresponding values were 5.1 events/h, 82.9%, 87.7%, 87.3%, and 0.45, respectively.
could accurately diagnose the sleep apnea patients based on the high accuracy and agreement of the results. Consequently, we can conclude that the results of our method were comparable to those of PSG. Because the PVDF film-based method is simple and does not require trained experts, if combined with signal processing unit, it can be used for ambulatory sleep apnea monitoring. Moreover, it can support apnea event detection during PSG recording.
The proposed unconstrained PVDF-based method had an accuracy similar to that for the ambulatory device that is currently used in clinical practice or constrained PVDF-based methods (see Table II). For instance, Ayas et al. used a wrist-worn device that combines a peripheral arterial tonometer (PAT), actigraph, and arterial oxygen saturation to diagnose OSA [19]. Even though the device used the attenuation of the PAT signal amplitude, which is strongly correlated with apnea [39], along with the SaO2 signal, which directly reflects the absence of breathing, the correlation coefficient between PSG AHI and the wrist-worn device was lower than that for ours. Another example is the NightWatch (NW) system, which records the SaO2 , nasal-oral airflow, and chest and abdominal wall motion using sensors attached to the patient’s face and body [20]. White et al. assessed the accuracy of the NW system used at home and in the lab to monitor sleep apnea. Although NW collected many physiological signals for apnea detection, its AHI estimation performance was similar to our results. However, in the present study, the method and system were not applied at home, and the apnea detection accuracy might be reduced under this condition. Han et al. also detected apneic events using a single nasal airflow channel from PSG data [38]. The performance of their method was found to be better than ours in a linear regression analysis of AHI, while the concordance on the Bland–Altman plot calculated in this study was slightly higher than that found in their study. Koo et al. detected respiratory events using PVDF impedance belts surrounding the patient’s chest and abdomen [31], and their method was comparable to standard respiratory inductance plethysmography in determining respiratory events during PSG. Nakano et al. monitored oronasal airflow using a PVDF thermal sensor, and found that an airflow monitor could be used to detect SDB [37]. Berry et al. compared the readings from a PVDF thermal sensor attached to the upper lip with a mask pneumotachograph and accurately detected respiratory events compared with the detection accuracy of pneumotachography in patients with OSA [30]. Even though they used belts or a thermal sensor that could directly measure the respiratory-induced signals, there were no significant differences in the AHI estimation results between the constrained methods and our unconstrained method. In particular, the apnea diagnostic ability at a fixed AHI threshold of the proposed method performed better in relation to the sensitivity, specificity, kappa statistics, and area
HWANG et al.: UNCONSTRAINED SLEEP APNEA MONITORING USING POLYVINYLIDENE FLUORIDE FILM-BASED SENSOR
under the ROC than the other methods with a PVDF-based sensor [31], [37]. Moreover, the greatest disadvantage of these systems was the necessity for the subject to wear or attach the PVDF-based sensor to their body or face during sleep, which could interrupt their normal sleep architecture. However, the number of analyzed apnea datasets in this study was less than those of the previous PVDF-based studies [31], [37], and our apnea detection performance could be different with larger data sets. C. Validation of PVDF Film-Based Sensors In this study, the PVDF sensors were uniformly oriented under the subject’s back position in a particular direction. To ensure the validity and reliability of the respiratory signal measurements, the PVDF sensors were aligned horizontally, as shown in Fig. 1, because subjects tend to move more often from side to side than in the transverse planar direction (up and down) during sleep. Sensors that are similar in shape and size to ours can easily be found in several commercialized sleep monitoring devices. For example, SleepScan SL-501 (TANITA, Tokyo, Japan) is a device that collects numerous kinds of physiological signals during sleep, including the pulse rate, respiration rate, and body motion data, and it is designed to be long in the horizontal direction. Other analogous examples are the EarlySense System (EarlySense, Waltham, MA, USA) and Nemuri (which means “Sleep”) SCAN NN-1300 (PARAMOUNT BED, Tokyo, Japan). Using PVDF sensors, we could obtain not only respiration signals, but also BCG signals. Moreover, when the PVDF sensor was aligned horizontally, the best signal-to-noise ratio (SNR) for the BCG signal was revealed in the preliminary test. It is speculated that blood ejection-induced vibration was transferred more strongly along a particular axis to the sensor through the bed mattress during the recording. However, the results of the SNR test that depended on the PVDF direction were not included in this paper. As mentioned in Section II, the data from PVDF channels #1 and #2 that reflected changes in the chest volume induced by breathing were not selected for the final analysis. From the training set data, the apnea detection performances using all of the channels were lower than those from channels #3 and #4. Because the subjects slept most of the time with pillows during PSG, there was a small gap between the bed mattress and a subject’s upper body. One could speculate that the volume change in the chest during respiration did not effectively transfer to the PVDF because the upper body was not fully in contact with the sensor. Furthermore, a PVDF impedance belt-based study showed that the respiratory signal measured by a chest belt was less correlated with the airflow than that from an abdomen belt [31]. As a result, channels #1 and #2 showed relatively high error components in the PVDF signals, which could degrade the apnea detection performance. This is why the data from these channels were finally excluded in the analysis. D. Validation of Apnea Detection Algorithm During PSG recording, the respiration patterns were different for the various subjects and varied depending on the position
2131
in bed and the sleep posture. To consider these things, adaptive thresholds were set based on the standard deviation of the PVDF signals for every minute for each subject in the apnea detection algorithm. As a result, we could obtain accurate AHI estimation results compared with PSG AHI. However, our method tended to underestimate apneic events at a higher AHI level, as shown in Fig. 4(a) and (b). At this level, apneic events occurred frequently in a short period, and the apneic event detection was ineffective under these conditions. In this study, the apnea detection algorithm was developed based on a “fixed time window” in consideration of real-time processing for ambulatory or home monitoring purposes. Apnea can persist for more than 30 s (1 epoch) in patients with severe apnea, and these durations can be diagnosed as normal breathing in our algorithm because the apnea threshold was set based on a change in the standard deviation of the PVDF data in the fixed time window. This is why we used a 60-s analysis window instead of 1 epoch, which is the most widely used time scale in sleep studies. Despite these efforts, the increased baseline of the threshold due to the restricted time window made it difficult to detect consecutive multiple apneic events. In contrast, our method tended to overestimate apneic events at the lower AHI level. One could speculate that our method may misestimate a respiratory signal drop as an apneic event that does not meet the standard scoring criteria. Moreover, other SRBD such as snoring, respiratory event-related arousals, upper airway obstruction, and Cheyne–Stokes breathing could occur in patients with sleep apnea [2]. These can also influence the amplitude decrease in the respiratory signals and might be estimated to be sleep apnea or hypopnea in our system. Moreover, PVDF signals may be distorted by motion artifacts that cannot be removed completely from the analysis process, which is one of the reasons that our method overestimates apneic events at a lower AHI. Despite these shortcomings, the overall apnea detection performance showed a high relationship between AHIs (R = 0.94). In addition, as shown in Table IV, the sensitivity was significantly lower than the specificity in the normal and mild severity groups. In these groups, the apnea detection sensitivity could be reduced considerably by only a few wrong estimates because the percentage of sleep apnea occurrences was markedly less than that of normal breathing. Thus, the kappa statistic was also used to evaluate the apnea detection performance. In particular, the difference between the sensitivity and specificity of the normal group was approximately 10% less than that of the moderate severity group, but there was no difference between the overall kappa statistics. Therefore, we could conclude that the proposed method showed similar apnea detection performances in all of the groups. In this study, the algorithm performed poorly when subjects slept in a lateral posture. During the respiratory cycle, respiratory-induced vertical (frontal axis) pressure was effectively transferred from the body to the PVDF sensors in a supine or prone posture. However, respiratory-related vertical pressure was transferred horizontally to the PVDF sensor in the lateral posture, and the respiratory signal measured by the sensor was significantly attenuated or distorted. As a result, the apnea detection performance in the lateral posture was
2132
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 7, JULY 2014
not validated for subjects who shared the bed with the partner. Additional PVDF sensor or increased size of sensor will probably be needed to detect sleep apnea for two people. Lastly, the practicability of the proposed method was not assessed in the home and ambulatory environments. In these environments, the apnea detection performance could be different, and problems that were not considered in the sleep laboratory might occur. To verify this issue, a validation of the proposed algorithm will be performed using PVDF signals measured in the home and ambulatory environments in a future study. V. CONCLUSION Fig. 6. PC1R P V D F , PC1R E S P , respiration from thoracic movement, respiration from nasal-oral airflow and apnea events, respectively. Amplitude of PC1R E S P is decreased with oscillation during OSA. On the other hand, amplitude of PC1R E S P is also decreased but without oscillation during CSA. In this figure, data from subject #29 were used.
relatively low and reflected a tradeoff between detection accuracy and unconstrained monitoring. In this study, very minute levels of oscillations such as “respiratory efforts” that can separate the OSA and central sleep apnea (CSA) could be measured using PVDF sensor, as shown in Fig. 6. In the figure, amplitude of respiration signal from nasal airflow sensor is decreased and that from chest belt sensor is decreased with oscillation during OSA. On the other hand, the amplitude from nasal airflow sensor has similar trend to OSA, but the amplitude of signal from chest belt sensor is decreased without oscillation during CSA. Those differences between OSA and CSA are also shown through our PVDF film-based system. So, we may think that OSA and CSA can be distinguished in our system using this amplitude decimation attenuation of the signal amplitude and oscillation occurrence. Although types of sleep apnea were not separated in this study, we plan to focus on OSA/CSA classification using PVDF film-based system to improve the impact of our work. E. Limitations Our proposed method also had some limitations. First, we did not separate sleep apnea into obstructive, central, and mixed types as mentioned in “Validation of apnea detection algorithm” session. With an improved algorithm and PVDF sensor, OSA and CSA can be divided into different types of apnea. Second, hypopnea events were not differentiated in the apneic events decision. However, this seems to have little clinical significance because both types of events have similar consequences and pathophysiologies [13]. In addition, AHI, which is an index of sleep apnea severity, is defined as the “sum” of apnea and hypopnea events instead of the frequency of each type of event [19]. Third, no analysis of the event-by-event apnea detection compared with PSG was conducted in this study because the coupled systems had different time scales for diagnosing sleep apnea. Fourth, test of the size and shape optimization of PVDF-based sensor was not performed because we were focusing on the possibility and potential of sleep apnea monitoring using PVDF film-based sensor in this study. Fifth, the proposed method was
In this study, a new sleep apnea monitoring method using a PVDF film-based sensor was developed that can unconstrainedly measure respiratory signals from subjects. Using the proposed system and method, we were able to obtain apnea detection performances similar to those of ambulatory devices and constrained PVDF-based studies. The developed system and method can be applied to sleep apnea detection for home or ambulatory monitoring purposes. REFERENCES [1] C. A. Kushida, M. R. Littner, T. Morgenthaler, C. A. Alessi, D. Bailey, J. Coleman, L. Friedman, M. Hirshkowitz, S. Kapen, M. Kramer, T. L. Chiong, D. L. Loube, J. Owens, J. P. Pancer, and M. Wise, “Practice parameters for the indications for polysomnography and related procedures: An update for 2005,” Sleep, vol. 28, no. 4, pp. 499–521, 2005. [2] M. H. Kryger, T. Roth, and W. C. Dement, “Normal human sleep: An overview,” in Principles and Practice of Sleep Medicine, 5th ed. Philadelphia, PA, USA: Saunders, 2011, pp. 16–26. [3] R. L. Wilkin, Clinical Assessment in Respiratory Care. St. Louis, MO, USA: Mosby, 1985, pp. 268–272. [4] B. El-Ad and P. Lavie, “Effect of sleep apnea on cognition and mood,” Int. Rev. Psychiatry, vol. 17, no. 4, pp. 277–282, 2005. [5] M. Hirshkowitz, I. Karacan, A. Gurakar, and R. L. Williams, “Hypertension, erectile dysfunction, and occult sleep apnea,” Sleep, vol. 12, no. 3, pp. 223–232, 1989. [6] S. Yan-fang and W. Yu-ping, “Sleep-disordered breathing: Impact on functional outcome of ischemic stroke patients,” Sleep Med., vol. 10, no. 7, pp. 717–719, 2009. [7] R. S. Leung, “Sleep-disordered breathing: Autonomic mechanisms and arrhythmias,” Prog. Cardiovasc. Dis., vol. 51, no. 4, pp. 324–338, 2009. [8] H. E. Resnick, S. Redline, E. Shahar, A. Gilpin, A. Newman, R. Walter, G. A. Ewy, B. V. Howard, and N. M. Punjabi, “Diabetes and sleep disturbances: Findings from the sleep heart health study,” Diabetes Care, vol. 26, no. 3, pp. 702–709, 2003. [9] Y. Peker, J. Hedner, A. Johansson, and M. Bende, “Reduced hospitalization with cardiovascular and pulmonary disease in obstructive sleep apnea patients on nasal CPAP treatment,” Sleep, vol. 20, no. 8, pp. 645–653, 1997. [10] T. Young, M. Palta, J. Dempsey, J. Skatrud, S. Weber, and S. Badr, “The occurrence of sleep-disordered breathing among middle-aged adults,” New Engl. J. Med., vol. 328, no. 17, pp. 1230–1235, 1993. [11] J. K. Kim, K. H. In, J. H. Kim, S. H. You, K. H. Kang, J. J. Shim, S. Y. Lee, J. B. Lee, S. G. Lee, C. Park, and C. Shin, “Prevalence of sleep-disordered breathing in middle-aged korean men and women,” Amer. J. Respir. Crit. Care. Med., vol. 170, no. 10, pp. 1108–1113, 2004. [12] P. E. Peppard, T. Young, J. H. Barnet, M. Palta, E. W. Hagen, and K. M. Hla, “Increased prevalence of sleep-disordered breathing in adults,” Amer. J. Epidemiol., vol. 177, no. 9, pp. 1006–1014, 2013. [13] “Sleep-related breathing disorders in adults: Recommendations for syndrome definition and measurement techniques in clinical research. The report of an American academy of sleep medicine task force,” Sleep, vol. 22, no. 5, pp. 667–689, 1999. [14] J. Boyle, N. Bidargaddi, A. Sarela, and M. Karunanithi, “Automatic detection of respiration rate from ambulatory single-lead ECG,” IEEE Trans. Inf. Technol. Biomed., vol. 13, no. 6, pp. 890–896, Nov. 2009.
HWANG et al.: UNCONSTRAINED SLEEP APNEA MONITORING USING POLYVINYLIDENE FLUORIDE FILM-BASED SENSOR
[15] M. O. Mendez, A. M. Bianchi, M. Matteucci, S. Cerutti, and T. Penzel, “Sleep apnea screening by autoregressive models from a single ECG lead,” IEEE Trans. Biomed. Eng., vol. 56, no. 12, pp. 2838–2850, Dec. 2009. [16] P. de Chazal, C. Heneghan, E. Sheridan, R. Reilly, P. Nolan, and M. O’Malley, “Automated processing of the single-lead electrocardiogram for the detection of obstructive sleep apnoea,” IEEE Trans. Biomed. Eng., vol. 50, no. 6, pp. 686–696, Jun. 2003. [17] F. Roche, J. M. Gaspoz, I. C. Fortune, P. Minini, V. Pichot, D. Duverney, F. Costes, J. R. Lacour, and J. C. Barth´el´emy, “Screening of obstructive sleep apnea syndrome by heart rate variability analysis,” Circulation, vol. 100, no. 13, pp. 1411–1415, 1999. [18] Y. G. Lim, K. H. Hong, K. K. Kim, J. H. Shin, S. M. Lee, G. S. Chung, H. J. Baek, D. U. Jeong, and K. S. Park, “Monitoring physiological signals using nonintrusive sensors installed in daily life equipment,” Biomed. Eng. Lett., vol. 1, pp. 11–20, 2011. [19] N. T. Ayas, S. Pittman, M. MacDonald, and D. P. White, “Assessment of a wrist-worn device in the detection of obstructive sleep apnea,” Sleep Med., vol. 4, no. 5, pp. 435–442, 2003. [20] D. P. White, T. J. Gibb, J. M. Wall, and P. R. Westbrook, “Assessment of accuracy and analysis time of a novel device to monitor sleep and breathing in the home,” Sleep, vol. 18, no. 2, pp. 115–126, 1995. [21] M. A. O’Reilly and K. Hynynen, “A PVDF receiver for ultrasound monitoring of transcranial focused ultrasound therapy,” IEEE Trans. Biomed. Eng., vol. 57, no. 9, pp. 2286–2294, Sep. 2010. [22] S. H. Hwang, H. N. Yoon, D. W. Jung, Y. J. Lee, D. U. Jeong, and K. S. Park, “Apnea event estimation during sleep using polyvinylidene fluoride film,” in Proc. eTELEMED, 2013, pp. 280–283. [23] J. Siivola, “New noninvasive piezoelectric transducer for recording of respiration, heart rate and body movements,” Med. Biol. Eng. Comput., vol. 27, no. 4, pp. 423–425, 1989. [24] G. R. Manjunatha, K. Rajanna, D. R. Mahapatra, M. M. Nayak, U. M. Krishnaswamy, and R. Srinivasa, “Polyvinylidene fluoride film based nasal sensor to monitor human respiration pattern: An initial clinical study,” J. Clin. Monit. Comput., vol. 27, pp. 647–657, Jun. 2013. [25] S. J. Choi and Z. W. Jiang, “A novel wearable sensor device with conductive fabric and PVDF film for monitoring cardiorespiratory signals,” Sens. Actuators, A, vol. 128, no. 2, pp. 317–326, 2006. [26] F. Wang, M. Tanaka, and S. Chonan, “Development of a PVDF piezopolymer sensor for unconstrained in-sleep cardiorespiratory monitoring,” J. Intel. Mater. Syst. Struct., vol. 14, no. 3, pp. 185–190, 2003. [27] H. J. Baek, G. S. Chung, K. K. Kim, and K. S. Park, “A smart health monitoring chair for nonintrusive measurement of biological signals,” IEEE Trans. Inf. Technol. Biomed., vol. 16, no. 1, pp. 150–158, Jan. 2012. [28] Y. Y. Chiu, W. Y. Lin, H. Y. Wang, S. B. Huang, and M. H. Wu, “Development of a piezoelectric polyvinylidene fluoride (PVDF) polymer-based sensor patch for simultaneous heartbeat and respiration monitoring,” Sens. Actuators A, vol. 189, pp. 328–334, 2013. [29] S. Redline, R. Budhiraja, V. Kapur, C. L. Marcus, J. H. Mateika, R. Mehra, S. Parthasarthy, V. K. Somers, K. P. Strohl, L. G. Sulit, D. Gozal, M. S. Wise, and S. F. Quan, “The scoring of respiratory events in sleep: Reliability and validity,” J. Clin. Sleep. Med., vol. 3, no. 2, pp. 169–200, 2007. [30] R. B. Berry, G. L. Koch, S. Trautz, and M. H. Wagner, “Comparison of respiratory event detection by a polyvinylidene fluoride film airflow sensor and a pneumotachograph in sleep apnea patients,” Chest, vol. 128, no. 3, pp. 1331–1338, 2005. [31] B. B. Koo, C. Drummond, S. Surovec, N. Johnson, S. A. Marvin, and S. Redline, “Validation of a polyvinylidene fluoride impedance sensor for respiratory event classification during polysomnography,” J. Clin. Sleep. Med., vol. 7, no. 5, pp. 479–485, 2011. [32] A. Rechtschaffen and A. Kales, A Manual of Standardized Terminology, Techniques and Scoring System for Sleep Stages of Human Subjects. Washington, DC, USA: NIH Publication, 1968. [33] R. O. Duda, P. E. Hart, and D. G. Stork, Pattern Classification, 2nd ed. Hoboken, NJ, USA: Wiley, 2001. [34] C. Iber, The AASM Manual for the Scoring of Sleep and Associated Events: Rules, Terminology and Technical Specifications. Westchester, IL, USA: Amer. Acad. Sleep Med., 2007, pp. 45–47. [35] J. M. Bland and D. G. Altman, “Statistical methods for assessing agreement between two methods of clinical measurement,” Lancet, vol. 1, no. 8476, pp. 307–310, 1986. [36] A. J. Viera and J. M. Garrett, “Understanding interobserver agreement: The kappa statistic,” Family Med., vol. 37, no. 5, pp. 360–363, 2005. [37] H. Nakano, T. Tanigawa, T. Furukawa, and S. Nishima, “Automatic detection of sleep-disordered breathing from a single-channel airflow record,” Eur. Respir. J., vol. 29, no. 4, pp. 728–736, 2007.
2133
[38] J. H. Han, H. B. Shin, D. U. Jeong, and K. S. Park, “Detection of apneic events from single channel nasal airflow using 2nd derivative method,” Comput. Meth. Programs Biomed., vol. 91, no. 3, pp. 199–207, 2008. [39] R. P. Schnall, A. Shlitner, J. Sheffy, R. Kedar, and P. Lavie, “Periodic, profound peripheral vasoconstriction—A new marker of obstructive sleep apnea,” Sleep, vol. 22, no. 7, pp. 939–946, 1999.
Su Hwan Hwang received the B.S. degree in biomechatronics and electronic electrical engineering from Sungkyunkwan University, Seoul, Korea, in 2009. He is currently working toward the combined M.Sc. and Ph.D. degrees in interdisciplinary program in bioengineering from Seoul National University, Seoul, Korea. His current research interests include biomedical signal measurement and processing in sleep medicine with the goal of translating the engineering works into practical medical research and clinical applications.
Hong Ji Lee received the B.S. degree in information communication engineering from Ewha Womans University, Seoul, Korea, in 2010. She is currently working toward the combined M.Sc. and Ph.D. degrees in interdisciplinary program in bioengineering from Seoul National University, Seoul, Korea. Her main research interests include the study of biomedical instrumentations and signal processing for ubiquitous healthcare. Her current research interests include tremor monitoring systems for diagnosis of Parkinson disease.
Hee Nam Yoon received the B.S. degree in biomedical engineering from Kyung Hee University, Yongin, Korea, in 2011. He is currently working toward the combined M.Sc. and Ph.D. degrees in interdisciplinary program in bioengineering from Seoul National University, Seoul, Korea. His current research interests include biomedical signals and systems for sleep medicine and synchronization dynamics of human biological signals.
Da Woon Jung received the B.S. degree in biomedical engineering from Kyung Hee University, Yongin, Korea, in 2011. She is currently working toward the combined M.Sc. and Ph.D. degrees in interdisciplinary program in bioengineering from Seoul National University, Seoul, Korea. Her current research interests include biomedical signals analysis for sleep medicine and synchronization dynamics of human biological signals.
2134
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 7, JULY 2014
Yu-Jin G. Lee received the M.D. degree from Soonchunhyang University Hospital, Cheonan, Korea, in 2005. She is a Psychiatrist and a Specialist in sleep medicine. She was a Clinical Fellow at the Center for Sleep and Chronobiology, Seoul National University Hospital. She is currently with the Seoul Metropolitan Eunpyeong Hospital, Seoul, Korea. Dr. Lee is a Member of the Board of Directors of the Korean Academy of Sleep Medicine.
Yu Jin Lee received the M.D. degree from Ewha Womans University, Seoul, Korea and the Ph.D. degree from the Chungbuk National University, Cheongju, Korea. She is currently an Associate Professor in the Division of Sleep and Chronobiology, Department of Psychiatry, College of Medicine and Hospital, Seoul National University Seoul, Korea.
Do-Un Jeong received the M.D. and Ph.D. degrees from Seoul National University, Seoul, Korea, in 1976 and 1988, respectively. He is currently a Professor of psychiatry at the Seoul National University College of Medicine and Director of the Center for Sleep and Chronobiology, Seoul National University Hospital. His research interests include signal processing in sleep and clinical sleep disorders. Dr. Jeong is the former President of the Korean Society of Medical and Biological Engineering and the Korean Academy of Sleep Medicine as well as a Fellow of the American Academy of Sleep Medicine.
Kwang Suk Park (M’78–SM’09) received the B.S., M.S., and Ph.D. degrees in electronics engineering from Seoul National University, Seoul, Korea, in 1980, 1982, and 1985, respectively. In 1985, he joined the Department of Biomedical Engineering, College of Medicine, Seoul National University, Seoul, as a Founding Staff Member, where he is currently a Professor. He is also the Director of Advanced Biometric Research Center, Seoul National University. His main research interests include biological signal measurement and processing for the diagnosis. His current research interests include nonintrusive measurements of biological signals for ubiquitous healthcare. Dr. Park is a member of the Korean Society of Medical and Biological Engineering and has served as the Secretary General of the World Congress on Medical Physics and Biomedical Engineering held in Seoul in 2006. He is also a Senior Member of the IEEE Engineering in Medicine and Biology Society. Since 2005, has been serving as an Associated Editor for the IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS. He has also chaired or co-chaired the Annual International Conference on u-Healthcare during past ten years.
