Wearable Pulse Oximetry Measurements on the Torso, Arms, and Legs: A Proof of Concept

Transcription

1 MILITARY MEDICINE, 182, 3/4:92, 2017 Wearable Pulse Oximetry Measurements on the Torso, Arms, and Legs: A Proof of Concept Marcus Kramer, PhD; Aaron Lobbestael, MS; Emily Barten, MS; John Eian, MS; Gregory Rausch, MS ABSTRACT For decades pulse oximeters designed for use on the head, hands, or feet have provided invaluable estimates of oxygen saturation to medical personal attending to combat casualties. However, traditional placement sites are not ideal for the relatively new paradigm of continuous battlefield telemonitoring. To assess the feasibility of oximetry on nontraditional body sites, 42 healthy volunteers were enrolled, consented, and underwent an industry standard induced-hypoxia study. During the study volunteers used prototype wearable oximeters, designed for the torso, arms, and legs. Subsets (size n) of the volunteers had the wearables placed at the following body sites, and achieved accuracies (A RMS, root-mean-square difference) of the following: calf 1.7% (n = 26); bicep 3.1% (n = 12); forearm 3.4% (n = 11); pectoral 2.9% (n = 42); sternum 2.9% (n = 13). In keeping with regulatory guidance calibrations with an A RMS of less than 3.5% are acceptable for potential future development. Additionally, a new method was developed to enable accurate reporting of respiration rate from the pectoral oximeter, A RMS of 1.1 breaths per minute (n = 10). This study demonstrates the feasibility of monitoring oxygen saturation and respiration rate from nontraditional sites via a wearable pulse oximeter. INTRODUCTION Recent years have seen an upsurge in the investigation of wearable sensors to meet the growing civilian and military needs for continuous critical care monitoring. Such wearable devices have included measures of heart rate (HR), temperature, oxygen saturation, electrocardiogram, and respiration rate among others. 1 In modern medical practice, peripheral oxygen saturation (SpO 2 ), HR, and respiratory rate have been identified as crucial metrics to protect against the onset of dangerous and potentially life-threatening conditions such as sepsis and pulmonary embolism. 2 Traditional pulse oximetry provides both SpO 2 and HR, whereas respiration rate can be provided through a variety of means including impedance pneumography or capnography. Still, the possibility of accessing all three of these parameters from a single sensor has led to many notable efforts to derive respiration rate from traditional pulse oximeters. 3 5 Likewise, there have been a number of attempts to incorporate oximetry into new wearable sensors. Unfortunately, these efforts have largely been limited to traditional oximetry sites. 6 For example, Mendelson et al has explored the implementation of a wearable oximeter on the forehead for military and civilian use. 7 Likewise, He et al described a new wearable oximeter for behind the ear, citing the location as having good pulse quality, being discrete and stable with natural anchoring to the ear. 8 Inevitably, to meet the growing medical needs of civilians and the military, new oximeter measurement technologies will need to be developed which can make a broader array of sites available. Nonin Medical, Inc., st Avenue North, Plymouth MN This article was presented at the Military Health System Research Symposium, Fort Lauderdale, FL, August 17 20, doi: /MILMED-D In traditional pulse oximetry the sensors are placed on the head, hands, or feet as a result of their strong perfusion, accessibility, and lower intersubject variability. Indeed, the fingers, palms, face, and ears have significantly higher perfusion than the rest of the body sites such as the torso, arms, and legs. 9,10 Likewise, with respect to physiologic characteristics such as adipose tissue, which accumulates disproportionally on the torso, legs, and upper arms, 11,12 the traditional sites have significantly lower intersubject variability. Similarly, the accessibility and reliability of transmission oximetry on the fingers, toes, and ears have made these sites a mainstay in modern medical practice. 6 Even forehead reflectance sensors have gained market acceptance as a replacement for the other traditional sites in cases of compromised circulation. 6,7 Nevertheless, soldiers on the battlefield require new solutions which free the head, hands, and feet, as these sites are often already in use or unavailable after injury. Notwithstanding the many advances in traditional pulse oximetry since its development in little progress has been made transitioning away from the traditional sites to new sites. 6 This lack of progress has hampered the development of new wearable medical devices which could incorporate these valuable parameters. Using our extensive experience designing small, lowpower reflectance pulse, and tissue oximetry systems 14 Nonin (Plymouth, Minnesota) has developed a wearable pulse oximeter for use on nontraditional sites. Presented here are the results of a pilot study to evaluate the accuracy of these new oximeters (Fig. 1A). The nontraditional body sites were selected from the torso, arms, and legs based on the technical feasibility, clinical acceptance, and access to the anatomical site. This new technology was assessed on a demographically diverse patient population in a standard-induced hypoxia clinical trial. 92

2 METHODS This study was approved by the University of California San Francisco Human Research Protection Program and conducted at the University of California San Francisco Hypoxia Research Laboratory in compliance with the principles of the Declaration of Helsinki. Study Design Forty-two healthy volunteers were enrolled, consented, and had sensors placed on up to five sites. The wearable sensors were placed on the chest, sternum, bicep, forearm, or calf before the start of hypoxia as shown in Figure 2. A Nonin signal processor (Model X-100SP) with a modified 8100AA (not available for sale) fingertip sensor, a commercially available fingertip oximeter (used by the research lab to monitor the patient) and a radial artery catheter were placed for monitoring during the study. Demographic (skin tone and race) and anthropometric (skin fold thickness measurements, height, and weight) information was collected before the study. Volunteers were recruited to reach an approximately equal mix of males and females with varying skin pigmentations. Race was self-identified and was collected to ensure that a range of ethnicities were captured. Skin color was judged by the study coordinator and was collected to examine effects of skin pigmentation on oximeter accuracy. Given the risks involved in induced hypoxia studies, the volunteers were limited to the young (ages years) and healthy (excluded for a body mass index [BMI] greater than 31, pregnancy, respiratory conditions, etc.). The volunteers underwent-induced hypoxia using an industry-standard protocol. 15,16 During the study, end-tidal O 2, and CO 2 were monitored to assess subjects metabolic condition and aid in controlling the depth of hypoxia. Hypoxia was induced by lowering the oxygen content of the air respired by the volunteers to achieve arterial oxygen saturation (SaO 2 ) blood draws ranging from 70 to 100%. Volunteers were taken through seven saturation plateaus (Fig. 3) to a minimum SpO 2 at around 70%. After the volunteer s saturation stabilized according to the fingertip pulse oximeter at each plateau arterial blood was collected concurrently with optical data from the oximeters, and evaluated with three ABL90 FLEX CO-oximeters (Radiometer, Brønshøj, Denmark). The average of the three CO-oximetry measurements was used as the truth value to determine the accuracy of the wearable oximeters. System Design Raw optical measurements were recorded using a custom data collection system consisting of a data storage and visualization system, Nonin signal processors (Model X-100SP), and custom wearable sensors. The sensors were designed to be mechanically optimized to the arms, legs, and chest (Fig. 1A). The sensor was attached to the patient through a flexible, medically safe, adhesive patch bonded to the sensor housing (Fig. 2). The spacing between the emitter and detector were optimized to a value between 3 and 20 millimeters to maximize the pulse signal, optical power, and accuracy of the sensor from each site (Fig. 1B). The emitter was composed of light emitting diodes (LED) with two wavelengths in the 600 to 1000 nanometer range, and the detector was a silicon photodiode. The sensors were connected to a commercially available Nonin signal processor. The signal processor was hardwired directly to a laptop through a proprietary universal serial bus converter. The laptop was used to store and visualize raw optical measurements from the signal processor. Data were stored on the laptop for future processing and calibration. FIGURE 1. (A) Image of wireless prototype wearable oximeter. (B) Drawing of a reflectance pulse oximeter sensor. (C) Rendering of path light takes through tissue in reflection oximetry. Drawings not meant to construe actual or relative sizes, dimensionality or construction of prototype wearable oximeter. Data Analysis Following the study, the raw optical measurements were processed using Nonin s proprietary signal processing algorithm for pulse identification, qualification, and quantification. For each qualified pulse, the ratio of pulse amplitudes 93

3 FIGURE 2. Shows representative images of the sensor placement locations for the pectoral, sternum, bicep and forearm sensors, respectively, from left to right. The arrow is located at the center of the sensor and provides directionality indications for its placement. The outline is of the adhesive patch that surrounds the sensor. was output. Analysis was completed in SAS 9.3 (SAS Institute, Cary, North Carolina) and Matlab (Mathworks, Natick, Massachusetts) to calibrate the sensors and provide a preliminary assessment of the oximetry system performance. For each site, a unique calibration was determined using the entire range of data collected from all subjects. Measurements of SaO 2, averaged from the three CO-oximeters, and the average ratio of pulse amplitudes during the blood draw for each of the wearable sensors, were paired. SpO 2 was calculated using a single a priori model common to pulse oximetry: SpO 2 ¼ A R þ B C R þ D where R is the average ratio of pulse amplitudes from two wavelengths of light during the blood draw, and A, B, C, andd were calibrated for each site to minimize the root-mean-square difference between SaO 2 and SpO 2 (A RMS ) using data across the entire range collected. A RMS is a representation of the combination of systematic and random error. It is defined as the accuracy root-mean-square FIGURE 3. Diagram of hypoxia study protocol. The gas mixture of respired air was monitored for the duration of the study. By adjusting the mixture of the delivered gas the peripheral oxygen saturation was systematically dropped to plateaus between 100% and 70%. Following hypoxia 100% O 2 was given to the volunteers to return their peripheral saturation to 100%. error difference between the measured value and the truth value calculated as: sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi X n ð i¼0 A RMS ¼ SpO 2iÞ 2 n where n is the number of observations, and both SpO 2 and SaO 2 are defined above. Observations were dropped in accordance with food and drug administration (FDA) standards for data analysis, 16 due to no good pulses being declared from Nonin s proprietary signal processing or because they were outside of the standard range for reporting accuracy of pulse oximeters (70 100%). The preliminary accuracy assessment, the mean bias is presented as the least squares mean for each site between 70% and 100% as estimated by a mixed model that accounts for the multiple measurements and multiple measurement sites per subject. 17 The Pearson correlation coefficient (PCC) is also presented by site for saturations between 70% and 100%. The 95% bias-corrected adjusted bootstrap confidence intervals are provided for the A RMS 18 using 2000 bootstrap samples. Summary statistics and data plots were also generated for comparison among the potential anatomical body sites. To compare between sites confidence intervals of the difference in A RMS between sites are provided. These intervals account for some subjects having more than one site tested at a time. Sites with an A RMS value of less than 3.5% for SaO 2 values between 70% and 100% are considered favorable for future development in keeping with regulatory guidance for reflectance oximeters. 19 A RMS statistics were also generated for the full range of data available to demonstrate and compare the accuracy of the device beyond the range used by the FDA to regulate 94

4 pulse oximetry. A RMS values are reported for the entire range, the 60% to 100% range and the 70% to 100% range in the manuscript. Assessment of the effect of demographic and anthropometric characteristics on accuracy was completed through analysis of variance for the categorical variables (gender, skin tone, and race), and generalized linear regression for continuous variables (height, weight, BMI, and blood pressure) with a p value of less than 0.05 signifying a statistically significant effect on bias after adjusting for using a false discovery rate adjustment for multiplicity. A proprietary digital signal processing algorithm was developed to estimate respiration rate from the pectoral sensor s optical data. The respiration rate algorithm was developed using data from subjects undergoing spontaneous breathing during the course of the hypoxia study described above. Paired capnography datawere available for 10 of the 42 hypoxia subjects, and portions of data from a subset of those subjects were used to develop the respiration rate algorithm. Once developed the algorithm was applied to rolling 30 second windows of pectoral data from all 10 subjects to provide a mean respiration rate over the windows. Quality metrics for the respiration rate algorithm were defined to be the correlation, bias, and A RMS between the algorithm respiration rate and a respiration rate inferred from the paired capnography data. RESULTS Forty-two healthy, nonsmoking volunteers were enrolled in the study and underwent hypoxia. Sensors were placed on all 42 volunteers but not all volunteers received the same sensors. See Table I for details on the number of subjects each sensor was tested on. The study sample had an average age of 26 years, with an average BMI of 23 kg/m 2 (range kg/m 2 ). Subjects were mainly Caucasian (27/42) and male (34/42) (see Table II for more details). Demographic and anthropomorphic characteristics were not found to have a significant impact on the accuracy of the tested devices as determined by the metrics set forth in the methods section. Data points from the study were dropped in accordance with the methods described previously. Table I reports the TABLE II. Demographic and Anthropomorphic Characteristics of Volunteers Characteristic Level Results Age 26 ± 5 Range: Gender Male 23/42 (55%) Height (Self-Reported) 69 ± 3 Range: Weight 153 ± 22 Range: BMI 23 ± 3 Range: Blood Pressure (mmhg) 133 ± 15 Range: Race White 27/42 (64%) Very Light 5/42 (12%) Light 22/42 (52%) Skin Tone Intermediate 6/42 (14%) Dark 5/42 (12%) Very Dark 4/42 (10%) Data are presented as mean ± standard deviation. number of expected data points per site, with the number of draws dropped for various reasons (missing co-oximeter value, device signal quality errors, saturation greater than or equal to 100, or saturation less than 70), and the accuracy in each of the 3 saturation ranges. Raw optical measurements recorded using the data collection system were postprocessed to identify a photoplethysmograph which was used to derive SpO 2. Sensor performance on the legs and arms varied between sites. The calf was found to have an accuracy, A RMS of 1.7 % (95% bootstrap confidence interval [1.5, 2.1], number of subjects = 26, mean bias = 0.17, PCC = 0.99, SaO 2 range = %) in line with the modified 8100AA fingertip oximeter, A RMS of 2.0% (95% bootstrap confidence interval [1.5, 3.5] number of subjects = 42). However, performance was significantly degraded (Tables I and III) on the bicep where the A RMS was 3.1% (95% bootstrap confidence interval [2.1, 5.6], number of subjects = 12, mean bias = 0.32, PCC = 0.95, SaO 2 range = %) or the forearm which had an A RMS of 3.4% (95% bootstrap confidence interval [2.4, 5.2], number of subjects = 11, mean bias = 0.48, PCC = TABLE I. Accuracy of Wearable Pulse Oximeters Used on Nontraditional Body Sites From Single Site Specific Calibration Across All Subjects Site Number of Subjects Draws Expected Draws Missing Cooximetry Draws With Signal Quality Erors Draws Used (60,101) Minimum SaO 2 Maximum SaO 2 A RMS With SaO 2 (60,101) Bicep (2.1, 5.6) Calf (1.5, 2.1) Forearm (2.4, 5.2) Pectoral (2.4, 3.5) Sternum (2.0, 4.8) 95

5 TABLE III. Comparison Between Accuracy (A RMS,70 100%) from Different Sites Site 1 Site2 Difference in A RMS Bicep Forearm 0.4 ( 1.4, 1.3) Calf Bicep 1.3 ( 3.9, 0.3) Calf Forearm 1.7 ( 3.5, 0.8) Calf Pectoral 1.2 ( 1.9, 0.7) Calf Sternum 1.1 ( 2.5, 0.2) Pectoral Bicep 0.1 ( 2.2, 0.7) Pectoral Forearm 0.5 ( 1.7, 0.4) Pectoral Sternum 0.1 ( 1.6, 1.3) Sternum Bicep 0.2 ( 2.3, 1.6) Sternum Forearm 0.6 ( 2.5, 1.0) 0.95, SaO 2 range = %). SpO 2 accuracy results are summarized in Table I. Similar results are seen for the sensors when examining their full range of data collected, 60% to 100%, which is outside of FDA reporting guidelines, 70% to 100% (Table I and Fig. 4). Even with the encumbrance of large motion from respiration, the sternum and pectoral oximeters both produced acceptable accuracies. During the study the pectoral oximeter was placed on 42 subjects with an accuracy of 2.9% (95% bootstrap confidence interval [2.4, 3.5], number of subjects = 42, mean bias = 0.12%, PCC = 0.96, SaO 2 range = %), whereas the sternum oximeter was only placed on 13 subjects but also produced an accuracy of 2.9% (95% bootstrap confidence interval [2.0, 4.8], number of subjects = 13, mean bias = 0.18, PCC = 0.96). Results on the torso are correspondingly worse than on the calf (Tables I and III) but still within guidelines set for the study. Similar results are seen for the sensors when examining their full range of data collected, 60% to 100%, which is outside of FDA reporting guidelines, 70% to 100% (Table I and Fig. 4). Further analysis of the pectoral sensor data also uncovered a new method to determine respiration rate. End-tidal CO 2 and oximeter data could be paired for 10 of the 42 subjects. The other 32 subjects either did not have end-tidal CO 2 recordings completed or the data could not be paired with the oximeter data. An accuracy, A RMS, of 1.1 breaths per minute, was achieved with a highly variable respiration rate between 10 and 45 breaths per minute (mean bias = 0.1%, PCC = , Fig. 5). DISCUSSION The development of a wearable pulse oximeter for nontraditional sites is a challenging endeavor. In traditional transmission pulse oximetry the LED and detector are placed on opposing sides of the tissue bed. The light emitted by the LED is predominantly scattered forward through the tissue, therefore, the majority of the light is transmitted to the opposing detector. In reflectance pulse oximetry, the detector is placed on the same side of the tissue as the LED (Fig. 1B) and a small portion of the light scattered by the tissue will instead reflect back to the same side of the tissue it came from (Fig. 1C). Therefore, only the small fraction of light reflected by the tissue will reach the detector and be usable for measurement. Methods such as driving the LEDs harder or an annular photodiode 7,20 could be used to produce an equivalent optical signal to that of transmission oximetry but require more power, or a larger sensor size. During the development of these wearable oximeters Nonin leveraged their experience designing medical grade oximeters that are favored 21 for their small size and low power. The current prototype oximeter incorporates Nonin s commercially available low-power and low-noise signal processor (model X-100SP) in combination with anatomically and optically optimized reusable sensors. However, through minor modification the signal processor cable can be removed to convert the unit into the wireless, wearable oximeter module shown in Figure 1A. This system works in combination with proprietary digital signal processing techniques providing a high-quality, low-noise optical measurement. The 10 blood draws per subject, expected for the calibration based on the study design, were not realized due to co-oximeter values missing, and signal quality errors. The raw optical measurements were digitally processed to improve signal quality, before calibration. This involves multiple stages of filtering and signal quality testing used to distinguish heart beats from noise or interfering signals. Inappropriate signals or distorted heartbeats are discarded by the signal quality tests. As a result, fewer blood draws are available for the TABLE I. Accuracy of Wearable Pulse Oximeters Used on Nontraditional Body Sites From Single Site Specific Calibration Across All Subjects Draws With SaO Draws With SaO 2 (60,100) A RMS With SaO 2 (60,100) Draws With SaO 2 <70 Draws With SaO 2 (70,100) Least Squares Mean Bias With SaO 2 (70,100) PCC A RMS With SaO 2 (70,100) (2.1, 6.2) ( 0.52, 1.17) (2.1, 5.6) (1.5, 2.2) ( 0.14, 0.47) (1.5, 2.1) (2.5, 5.4) ( 0.27, 1.23) (2.4, 5.2) (2.3, 3.6) ( 0.41, 0.16) (2.4, 3.5) (2.1, 5.0) ( 0.54, 0.90) (2.0, 4.8) 96

6 FIGURE 4. This figure shows the arterial oxygen saturation (SaO 2 ) plotted against the peripheral oxygen saturation with a calibration fit. The calibration produces an A RMS for the prototype bicep 3.2% (n = 12, %), calf 1.8% (n = 26, %), forearm 3.6% (n = 11), pectoral 2.8% (n = 42) and sternum 3.0% (n = 13) wearable oximeters (60-100%). FIGURE 5. Shows the respiration rate calibration data from the pectoral wearable oximeter with an accuracy of 1.1 breaths-per-minute. The end-tidal CO 2 respiration rate, as estimated from a capnograph, is plotted against the optically derived respiration rate. calibration model used to assess the accuracy (A RMS ) of the oximeter system. This can be seen in Table I where, e.g., the pectoral sensor should have 420 blood draws but only 389 were used in the calibration. In accordance with regulatory standards a minimum of 200 data points from at least 10 subjects, and A RMS better than 3.5% between 70% to 100% are required for a reflectance oximeter to be cleared for medical use in the United States. However, due to the inaccuracy of the reference pulse oximeter (and co-oximeter) and challenges in precisely controlling the desaturation between subjects; the true saturation (SaO 2, as measured from the co-oximeter) can inadvertently go outside these guidelines. Therefore, we have reported results for all sensors based on the 70% to 100% criterion to provide a direct comparison of this study s A RMS to those reported by medical device manufacturers for the fingers, ears, and forehead (Table I). Of the sites tested in this study, the upper pectoral and calf both were better than the minimum FDA requirement. As well, the reported accuracy on the calf is equivalent with commercially available fingertip pulse oximeter accuracies of ± 2% (A RMS ) and superior to all other sites reported in this study (Table III). It should be recognized that the FDA rounds accuracies to the nearest whole number so both the 1.7% (calf) and 2.0% (8100AA fingertip oximeter) accuracies would be reported as an accuracy of ± 2% by a medical device manufacturer. These sites are therefore very promising for further development. Whereas, the forearm, bicep, and sternum lack enough data points, they do meet the basic accuracy requirements (between 70% and 100%) for further development. Today the standard of care for patients uses the finger-tip pulse oximeter that monitors SpO 2 and HR. However, in their comparative study on key contributors of death, Lynn et al showed that these metrics alone do not provide a holistic picture of patient health. 2 Instead, with the addition of respiration rate, these three metrics may serve to minimize the occurrences of sepsis and pulmonary embolism while aiding in the detection of respiratory distress or failure, assisting in triage efforts. This new wearable oximeter technology could potentially provide SpO 2, HR, and respiration rate to military and civilian medical personnel, delivering essential metrics during critical care. LIMITATIONS This study was limited to young healthy adults. The devices described are prototypes that have not been cleared by the FDA. This study did not include an appropriate truth (e.g. electrocardiography) for the HR parameter derived from the wearable oximeters and as such, accuracy was not presented 97

7 for the parameter. Likewise, the results of this study only estimate the accuracy of SpO 2 and respiration rate not the overall performance of these devices across patient populations and use cases. Furthermore, because these data were used to calibrate the oximeter, we anticipate further study is required to validate the device accuracy. CONCLUSIONS This study demonstrates the feasibility of accurately monitoring oxygen saturation from the torso, arms, and legs via a miniaturized wearable pulse oximeter system which combines low-power and low noise hardware with cutting-edge signal processing. The unique calibration, patient interface, and sensor design are some of the many reasons reflectance sensors work effectively only on sites to which they have been optimized. These newly developed wearable oximeters have the potential to perform across nontraditional body sites enabling the monitoring of wounded warriors during the continuum of care. Additionally, the torso oximeter has been enabled to improve critical casualty care by adding accurate respiration rate to its parameter set. Further development will focus on consolidating the design into a single wearable module, validating the accuracy of this new wearable oximetry system and expanding the performance across use cases. This technological evolution represents the first step of developing wearable oximeters for use during the continuum of battle and critical care. ACKNOWLEDGMENTS This study was funded by Nonin Medical, Inc. as part of ongoing research and development. REFERENCES 1. Pantelopoulos A, Bourbakis NG: A survey on wearable sensor-based systems for health monitoring and prognosis. Trans Sys Man Cyber Part C 2010; 40(1): Lynn L, Curry J: Patterns of unexpected in-hospital deaths: a root cause analysis. Patient Saf Surg 2011; 5(1): Shelley KH, Jablonka DH, Awad AA, Stout RG, Rezkanna H, Silverman DG: What is the best site for measuring the effect of ventilation on the pulse oximeter waveform? Anesth Analg 2006; 103(2): Lazaro J, Gil E, Bailon R, Laguna P: Deriving respiration from the pulse photoplethysmographic signal. Comput Cardiol 2011; 38(9): Orini M, Pelaez-Coca MD, Bailon R, Gil E: Estimation of spontaneous respiratory rate from photoplethysmography by cross time-frequency analysis. Comput Cardiol 2011, 38(9): Tamura T, Maeda Y, Sekine M, Yoshida M: Wearable photoplethysmographic sensors past and present. Electronics 2014; 3(2): Mendelson Y, Duckworth RJ, Comtois G: A wearable reflectance pulse oximeter for remote physiological monitoring. Conf Proc IEEE Eng Med Biol Soc. 2006; 1: Available at: search/searchresult.jsp?newsearch=true&querytext=mendelson%20y, %20Duckworth%20RJ,%20Comtois%20G.%20A%20wearable%20reflec tance%20pulse%20oximeter%20for%20remote%20physiological%20moni toring.%20%20proceedings%20of%20the%2028th%20ieee; accessed April 5, He DD, Winokur ES, Heldt T, Sodini CG: The ear as a location for wearable vital signs monitoring. Conf Proc IEEE Eng Med Biol Soc. 2010; Available at: xpl/articledetails.jsp?arnumber= ; accessed February 22, Harbi P, Thacher T: Body mapping of human cutaneous microcirculatory perfusion using a real-time laser Doppler imager. Diab Vasc Dis Res 2013; 10(2): Tur E, Tur M, Maibach HI, Guy RH: Basal perfusion of the cutaneous microcirculation: Measurements as a function of anatomic position. J Investig Dermatol 1983; 81(5): McArdle WD, Katch FI, Katch VL: Exercise Physiology: Energy, Nutrition, and Human Performance. Philadelphia, PA, Lippincott Williams & Wilkins, Ross R, Shaw KD, Martel Y, de Guise J, Avruch L: Adipose tissue distribution measured by magnetic resonance imaging in obese women. Am J Clin Nutr 1993; 57(4): Severinghaus JW: Takuo Aoyagi: discovery of pulse oximetry. Anesth Analg 2007; 105(Suppl 6): S MacLeod DB, Ikeda K, Vacchiano C, Lobbestael A, Wahr JA, Shaw AD: Development and validation of a cerebral oximeter capable of absolute accuracy. J Cardiothorac Vasc Anesth 2012; 26(6): ISO: Clinical investigation of medical devices for human subjects Good clinical practice. International Organization for Standardization. Available at accessed February 22, ISO: Medical electrical equipment Part 2-61: Particular requirements for basic safety and essential performance of pulse oximeter equipment. International Organization for Standardization. Available at en; accessed February 22, Carstensen B: In Comparing Clinical Measurement Methods. John Wiley & Sons, Ltd, Efron B, Tibshirani RJ: An Introduction to the Bootstrap. Boca Raton, FL, CRC Press, Food and Drug Administration: Pulse Oximeters Premarket Notification Submissions [510(k)s]: Guidance for Industry and Food and Drug Administration Staff. Available at DeviceRegulationandGuidance/GuidanceDocuments/ucm htm; accessed February 22, Haahr RG, Duun SB, Toft MH, et al: An electronic patch for wearable health monitoring by reflectance pulse oximetry. IEEE Trans Biomed Circuits Syst 2012; 6(1): Tura A, Badanai M, Longo D, Quareni L: A medical wearable device with wireless bluetooth-based data transmission. Meas Sci Rev 2003; 3(2): 4. 98