A Wearable Vital Signs Monitor at the Ear for Continuous Heart Rate and Pulse Transit Time Measurements

Abstract

A continuous, wearable and wireless vital signs monitor at the ear is demonstrated. The device has the form factor of a hearing aid and is wirelessly connected to a PC for data recording and analysis. The device monitors the electrocardiogram (ECG) in a single lead configuration, the ballistocardiogram (BCG) with a MEMS triaxial accelerometer, and the photoplethysmograms (PPG) with 660nm and 940nm LED sources and a static photocurrent subtraction analog front end. Clinical tests are conducted, including Valsalva and head-up tilt maneuvers. Peak timing intervals between the ECG, BCG and PPG are extracted and are shown to relate to pre-ejection period and mean arterial blood pressure (MAP). Pulse Transit Time (PTT) extracted from cross-correlation between the PPG and BCG shows improved results compared to the pulse arrival time (PAT) method for tracking changes in MAP.

Full text


Abstract²A continuous, wearable and wireless vital signs
monitor at the ear is demonstrated. The device has the form
factor of a hearing aid and is wirelessly connected to a PC for
data recording and analysis. The device monitors the
electrocardiogram (ECG) in a single lead configuration, the
ballistocardiogram (BCG) with a MEMS triaxial
accelerometer, and the photoplethysmograms (PPG) with
660nm and 940nm LED sources and a static photocurrent
subtraction analog front end.
Clinical tests are conducted, including Valsalva and head-up
tilt maneuvers. Peak timing intervals between the ECG, BCG
and PPG are extracted and are shown to relate to pre-ejection
period and mean arterial blood pressure (MAP). Pulse Transit
Time (PTT) extracted from cross-correlation between the PPG
and BCG shows improved results compared to the pulse arrival
time (PAT) method for tracking changes in MAP.
I. INTRODUCTION
Cardiovascular disease (CVD) affects more than 80
million people as of 2008 and is the leading cause of death in
the U.S. [1]. In 2008, costs associated with CVD were
$297.7 billion, and by 2030, costs are expected to reach
$1.117 trillion per year for CVD in the U.S. alone [1]. To
help reduce these costs, there is a push to change the current
hospital-centric, reactive healthcare delivery system to one
that focuses on early detection and diagnosis through
extended, personalized monitoring [2] [3].
Continuously monitoring vital signs such as heart rate
(HR) and heart intervals can provide the data necessary for
early diagnosis of CVD [4]. Several groups have designed
measurement systems to monitor many of these vital signs
for personalized health care. Guo et al. designed a wearable
vital signs monitor with a chest belt for electrocardiography
(ECG) and an ear-worn probe for photoplethysmography
(PPG) [5]. They measured heart rate, peripheral oxygen
saturation (SpO2) and systolic blood pressure using the
pulse-arrival time (PAT) method. Pinheiro et al. measured
the PPG at the finger, the ECG at the chest and the
ballistocardiogram (BCG) from a chair for heart rate
variability (HRV), HR, PAT and SpO2 measurements [6].
Here, we demonstrate a continuous, wearable and
wireless vital-sign monitor worn at the ear in the form factor
of a hearing aid. The ear location is chosen because it is a
This work was funded by the Medical Electronic Device Realization
Center (MEDRC) at MIT. Human subject testing was conducted at the MIT
Catalyst Clinical Research Center (CRC) under IRB approval
#1104004449. The CRC is supported through grant #UL1 RR025758 from
the NIH through the National Center for Research Resources.
E. S. Winokur, D. He and C. G. Sodini are all with the Massachusetts
Institute of Technology, Cambridge, MA 02139 USA. (e-mail:
ewinokur@mit.edu, davidhe@mit.edu, sodini@mtl.mit.edu)
natural anchoring point, and it is discreet since the device
can be partially hidden by hair and the ear [7]. It is also
chosen so that all physiological signals can be acquired from
one location for improved patient compliance. The monitor
measures single lead ECG, BCG, and PPG. These
physiological signals allow the monitor to measure heart rate
from three independent sources, and blood oxygenation.
Additionally, information embedded in the peak timings
between the signals has been shown to correlate to the pre-
ejection period (PEP), heart contractility, stroke volume,
cardiac output and mean arterial blood pressure, the last of
which will be discussed in this paper [8 ± 11].
II. MEASURING PHYSIOLOGICAL SIGNALS AT THE HEAD
A. Head Ballistocardiogram (BCG)
The BCG is a measure of thHERG\¶VPHFhanical reaction
to the blood ejected during systole and is traditionally
measured by having a subject lie on a low-friction bed [8].
As noted in [9], the head BCG recoil is in the head-to-foot
axis, which is consistent with the principal blood-volume
shift during cardiac ejection. These movements are typically
10 mGp-p. To measure the BCG, we employ a Bosch
BMA180 MEMS triaxial accelerometer with a 9 Hz
bandwidth, 14 bit resolution, and ± 2G range [9]. The low
SNR of the BCG signal requires this measurement to be
taken at rest.
B. Head Electrocardiogram (ECG)
Due to the conductive nature of the body, the ECG can
be measured in an attenuated form at different locations
away from the heart. A single-lead ECG is used, with one
electrode placed on the mastoid bone, and the other on the
neck. The ECG front-end circuit is shown in Fig. 1 [9]. The
instrumentation amplifier (IA) has an adjustable gain to
amplify QRS complexes on the order of 30 ± 40Vp-p. The
low SNR of the ECG signal requires this measurement to be
taken at rest.
Fig. 1 The head ECG front end circuit. IA: instrumentation amplifier; ADC:
analog-to-digital converter.
C. Head Photoplethysmogram (PPG)
The head PPG is taken in reflectance mode, with light
reflecting off the mastoid bone to photodiodes in the same
A Wearable Vital Signs Monitor at the Ear for Continuous Heart
Rate and Pulse Transit Time Measurements
Eric S. Winokur, Student Member, IEEE, David Da He, Student Member, IEEE, and Charles G.
Sodini, Fellow, IEEE
34th Annual International Conference of the IEEE EMBS
San Diego, California USA, 28 August - 1 September, 2012
2724978-1-4577-1787-1/12/$26.00 ©2012 IEEE

plane as the LED light sources. The PPG has a small
pulsatile component, usually between 0.25 ± 1% of a large
static component [12]. This large static current presents
many challenges for low-voltage, single-supply systems.
Typically, reducing the LED drive current can maintain the
output of the transimpedance amplifier (TIA) within the
supply rails, at the expense of lower signal-to-noise ratio
(SNR). Alternatively, reducing the transimpedance of the
feedback element will achieve the same result.
Wong et al. developed a static photocurrent subtraction
circuit by using an error amplifier in feedback with the
transimpedance amplifier to drive a current source into the
cathode of the photo-diodes [13]. However, the frequency
response of the TIA changes due to the amount of static
photocurrent subtracted, and can reduce the bandwidth to
unacceptably low levels. Tavakoli utilized a log-amp in the
front end to increase dynamic range, requiring a full signal
path for each PPG wavelength [12].
Fig. 2 PPG front end with forward biased photo-diodes. The forward diode
current IF subtracts a static amount of photocurrent from Iph, which
increases the dynamic range of the front end.
To address these concerns, a new photoreceptor circuit
has been designed, in which the photodiodes are forward
biased (Fig. 2). A digital-to-analog converter (DAC) is used
to generate a reference voltage Vref at the positive terminal of
the TIA. The amount of static photocurrent subtracted from
the PPG is determined by the diode equation:
¸
¸
¹
·
¨
¨
©
§ 1
T
D
V
V
sFeII
(1)
where IF is the forward diode current, Is is the reverse bias
saturation current, VD is the voltage across the diode and VT
is the thermal voltage kT/q. Due to the virtual ground of the
TIA op-amp, Vref is equal to VD and therefore sets the
forward biased diode current. The photocurrent Iph, which
flows in the reverse direction of IF, is set through feedback
from the microcontroller unit (MCU) to the LED driver, so
that the remaining current IOA is sufficiently small for the op-
amp to supply. $FFRUGLQJWR.LUFKKRII¶V&XUUHQW /DZ
(KCL), the current into a node, must equal the current out:
phFOA III 
(2)
F
DD
OA
F
ref
R
V
I
R
Vdd

(3)
The exponential nature of (1) allows the total amount of
subtracted photocurrent to be large as given by (2), with the
current steps being defined by the resolution of the DAC. A
16-bit R-2R ladder DAC (Analog Devices AD5541A) has
sufficient resolution for this application and was used in this
implementation.
Feedback from the MCU (Texas Instruments MSP430) is
used to control the LED drive currents and DAC voltage. If
the LED drive current becomes too large, the MCU can
lower Vref to reduce power consumption of the system. Fig. 3
shows the ECG, BCG and PPG (red light only) signals
simultaneously measured at the head. LED optical output
power during these experiments was limited to between
0.027 and 0.068 lumens.
Fig. 3 ECG, BCG, and PPG data collected simultaneously at the head
III. SYSTEM DESCRIPTION
The system consists of the BCG accelerometer, the ECG
front end, the PPG front end, a power management circuit, a
microcontroller, a wireless transmitter and receiver, the PC
software and the mechanical housing.
The BCG, ECG and PPG components have been
described in Section II. Power is supplied by a 3V lithium
coin cell battery (Panasonic CR2032) and the rest of the
power management circuitry is described in [9].
The MCU shuttles the continuous real time BCG, ECG
and PPG data to the 2.4 GHz low-power wireless transmitter
(Texas Instruments CC2500). The raw data is displayed in
MATLAB in real time for visual feedback during testing.
Fig. 4 Left: The wearable vital signs monitor worn at the ear. Center top:
Front side of the monitor with MSP430, radio, battery and PPG sensor
outlined. Center bottom: Back side of the monitor with PPG and ECG front
ends, power management and accelerometer outlined. Right: The USB
interface that allows the device to interface with MATLAB.
The mechanical housing of the device is modified from a
hearing aid housing, which consists of a plastic shell that is
attached to an ear-bud. The custom designed Rigid ± FLEX
PCB is 0.5 mm thick and contains a 5.08cm flexible
connector for the PPG subsystem to fold over and interface
with the mastoid bone. The PPG sensor has 660nm and
940nm LED sources surrounded by 4 photodiodes
(Advanced Photonix PD-C160SM). Fig 4 shows the entire
prototype system anchored to the ear.
2725

IV. DETERMINING HEART INTERVALS USING
SIMULTANEOUS ECG, BCG, AND PPG
Several heart intervals can be determined via the three
physiological signals measured with the vital signs monitor;
Pre-ejection period (PEP), defined as the electro-mechanical
delay between the onset of (&*¶V4ZDYH and the time when
the aortic valve opens [11]; pulse arrival time (PAT), defined
as the time between the Q wave and the arrival of the pulse
wave to a location on the body [11]; pulse transit time
(PTT), defined as the time it takes for a pulse wave to travel
from one location to another location in the vasculature [11].
A. ECG & BCG: PEP
The vital-signs monitor in practice estimates the PEP as
the delay between the R wave of the ECG, and the principal
wave of the BCG (the J wave). This measurement includes
the PEP, as well as a small portion of PTT, as the J wave
motion is due to the blood pulse moving around the aortic
arch and down the descending aorta [9]. The PEP contains
information related to the contractility of the heart, stroke
volume, cardiac output, and autonomic responses [10].
B. ECG & PPG: PAT
PAT is measured in practice as the delay between the R
wave of the ECG and diastole of the PPG [14]. PAT includes
both PEP and PTT intervals. Information from the PAT has
been used to estimate cuff-less mean arterial blood pressure
(MAP). However many inaccuracies may arise due to the
changes in PEP, which are not necessarily correlated with
MAP [14]. This will be discussed further in Section V.
C. PAT ± PEP = PTT
When subtracting PEP from PAT, the remaining signal is
purely PTT. MAP has been empirically related to PTT via a
negative logarithmic relationship through a long tube pulse
wave velocity model [15] [16]:
hydro
PPTTBAMAP u )ln(
(4)
In Eq. (4), A is a constant dependent on the elastic properties
of the vasculature and the distance the pulse wave has
traveled. B is also a constant dependent on elastic properties
of the vasculature and Phydro is the hydrostatic pressure due to
gravity [15][16]. Phydro becomes zero when the subject is
supine. Compared to measuring PTT from a finger PPG,
calibration for the vital signs monitor should be greatly
reduced since the head is almost always upright or level
compared to the heart [5]. The ability to subtract PEP
accurately from this equation removes one of the large error
sources associated with transit-derived ABP [11] [14].
V. MEASURED CHANGES IN HEART INTERVALS DUE TO
HEMODYNAMIC MANEUVERS
Fig. 5 shows continuous blood pressure from a Finapres
monitor, as well as -ln(PTT), R-J interval and -ln(PAT)
derived from signals measured simultaneously from the vital
signs monitor. All signals were manually annotated to ensure
accuracy.
Fig. 5 Top panel: Finapres continuous BP data before, during, and after a
Valsalva maneuver. The bold line corresponds to MAP. 2nd panel: -ln(PTT),
averaged over 5 beats showing qualitative correlation to MAP. 3rd panel: R-
J Interval during the maneuver averaged over 5 beats. 4th panel: -ln(PAT),
averaged over 5 beats not showing significant qualitative correlation to
MAP due to the change in PEP moving in the opposite direction as PTT.
The shaded regions correspond to the Valsalva maneuver.
Fig. 6 Finapres MAP and PTT derived MAP over a Valsalva maneuver.
The gray shaded region corresponds to Valsalva strain. Mean error between
the two waveforms is -0.07mmHg with a standard deviation of 3.64 mmHg.
During the intervention, the subject was at rest for 50s,
and then performed a Valsalva for approximately the next
20s, and finally was at rest again for the next 50s. The
change in ABP observed during the Valsalva maneuver
corresponds to the changes noted in [17]. A qualitative
correlation between Finapres MAP and -ln(PTT)) is
discernible, whereas no such qualitative correlation is seen
between MAP and -ln(PAT). The lack of qualitative
correlation from -ln(PAT) is due to the change in PEP during
the maneuver, moving in the opposite direction as PTT. As
the maneuver begins, PEP is shown to lengthen, whereas
after a brief lengthening of PTT, the autonomic response
causes vasoconstriction, thus reducing the PTT interval.
Fig. 6 compares transit derived MAP and Finapres MAP.
Solving for A and B in (4) yields 44 and 0.05 respectively.
2726

The mean error is -0.07 mmHg with a standard deviation of
3.64 mmHg.
Fig. 7 shows simultaneously measured PTT, PAT and R-J
intervals during a head-down tilt maneuver (from 75o head-
up tilt to supine). The PTT is measured by cross-correlating
and averaging the filtered BCG and derivative PPG; the PAT
is measured by cross-correlating and averaging the filtered
ECG and the derivative PPG; the R-J interval is measured by
cross-correlating and averaging the filtered ECG and BCG
waveforms [9]. During the first portion of the measurement,
the subject is tilted at 75o. After three minutes, the subject is
lowered to supine for another three minutes. The decrease in
R-J interval is caused by the influx of venous return as the
body goes to the supine position [10] [18]. The increase in
-ln(PTT) and -ln(PAT) is due to the increase in circulatory
blood as venous pooling is no longer occurring, and
qualitatively correlates to expected changes in blood
pressure [19] [20]. Here, -ln(PAT) correlates to the expected
change in blood pressure because the PEP and PTT intervals
are both changing in the same direction (decreasing).
Fig. 7 ±ln(PTT), -ln(PAT), and R-J interval measured from the vital signs
monitor during a tilt to supine maneuver.
VI. CONCLUSIONS AND FUTURE WORK
A wearable and wireless vital-signs monitor at the ear has
been demonstrated. The portable device measures the ECG,
BCG and PPG, all from a single area behind the ear. R-J
interval and ±ln(PTT) have been shown to correlate to pre-
ejection period and arterial blood pressure, respectively.
Future work to improve the power consumption and
accuracy of the measurements will occur by changing the
radio to BlueTooth Low Energy, designing custom
integrated circuits for the ECG and PPG front ends, and
improving the mechanical design of the device.
ACKNOWLEDGMENT
The authors would like to thank Dr. Thomas Heldt (MIT)
for his valuable discussions DQG7RP2¶'Z\HU(Analog
Devices) for his insight. The authors would also like to thank
Catherine Ricciardi and Ilene Horvitz for assisting in the
clinical tests and oversight for research subjects.
REFERENCES
[1] 5 / 9HURQLTXHHWDO³+HDUW'LVHDVHDQG6WURNH6WDWLVWLFV ± 2012
Update: A Report from WKH$PHULFDQ+HDUW$VVRFLDWLRQ´
Circulation, December, 2011.
[2] *7URVWHU³7KH$JHQGDRI:HDUDEOH+HDOWKFDUH´IMIA Yearbook of
Med. Info. 2005: Ubiquitous Health Care Systems, pp. 125-138,
2004.
[3] X.-F. Teng, Y.-7=KDQJ & & < 3RRQ DQG3 %RQDWR ³:HDUDEOH
Medical Systems for p-+HDOWK´IEEE Reviews in Biomedical
Engineering, vol. 1, pp. 62-74, 2008.
[4] $ /\PEHULV ³6PDUW :HDUDEle Systems for Personalised Health
0DQDJHPHQW&XUUHQWDQG)XWXUH&KDOOHQJHV´IEEE EMBC., pp.
3716-3719, 2003.
[5] D. G. Guo, F. E. H. Tay, L. Xu, M. N. Nyan, F. W. Chong, K. L. Yap
DQG% ;X ³$/RQJ-term Wearable Vital Signs Monitoring System
using BSN´ 11th EUROMICRO Conference on Digital System
Design Architectures, Methods and Tools, pp.825-830, 3-5 Sept.
2008
[6] E. Pinheiro, O. Postolache, and P. Girao, "Pulse arrival time and
ballistocardiogram application to blood pressure variability
estimation," IEEE International Workshop on Medical Measurements
and Applications, pp.132-136, 29-30 May 2009
[7] D. He, E. S. Winokur, T. Heldt and & * 6RGLQL ³7KH(DUDV D
/RFDWLRQIRU:HDUDEOH9LWDO6LJQV 0RQLWRULQJ´IEEE EMBC., pp.
6389-6392, 2010.
[8] I. Starr, A. J. Rawson+$6FKURHGHUDQG15-RVHSK³6WXGLHVRQ
the Estimation of Cardiac Output in Man, and of Abnormalities in
&DUGLDF)XQFWLRQ IURP WKH+HDUW¶V5HFRLODQGWKH%ORRG¶V ,PSDFWV
WKH%DOOLVWRFDUGLRJUDP´The American Journal of Physiology, vol.
127, pp. 1-28, 1939.
[9] D. He, E. S. Winokur and &* 6RGLQL ³$&RQWLQXRXV :HDUDEOH
and Wireless Heart Monitor using Head Ballistocardiogram (BCG)
DQG+HDG(OHFWURFDUGLRJUDP(&*´IEEE EMBC., pp. 4739-4732,
2011.
[10] M. Etemadi, O. T. Inan, R. M. Wiard, G. T. A. Kovacs, and L.
*LRYDQJUDQGL³1RQ-Invasive Assessment of Cardiac Contractility on
D:HLJKLQJ6FDOH´IEEE EMBC., pp. 6773-6776, 2009.
[11] G. Zhang, D. Xu, N. B. Olivier and R. Mukkamala, "Pulse arrival
time is not an adequate surrogate for pulse transit time in terms of
tracking diastolic pressure," IEEE EMBC, pp.6462-6464, 2011.
[12] M. D. TavDNROL ³$Q$QDORJ9/6,)URQW(QGIRU3XOVH2[LPHWU\´
PhD Thesis: MIT, Cambridge, MA, 2006.
[13] A. K. Y. Wong, K.-P Pun, Y.-T. Zhang DQG.1 /HXQJ³$/RZ-
Power CMOS Front-End for Photoplethysmographic Signal
Acquisition wLWK5REXVW'&3KRWRFXUUHQW5HMHFWLRQ´IEEE Trans.
On Biomed. Cir. And Sys., vol. 2 no. 4, pp. 280-288, December, 2008.
[14] P. Fung, G. Dumont, C. Ries, C. Mott, and M. Ansermino,
³&RQWLQXRXV 1RQLQYDVLYH%ORRG3UHVVXUe Measurement by Pulse
7UDQVLW7LPH´IEEE EMBC., pp. 738-741, 2004.
[15] : &KHQHWDO ³&RQWLQXRXV (VWLPDWLRQRI6\VWROLF%ORRG3UHVVXUH
8VLQJWKH3XOVH$UULYDO7LPHDQG,QWHUPLWWHQW&DOLEUDWLRQ´IEEE
Comput. Med. Biol. Eng., vol. 38, pp. 569 ± 574, 2000.
[16] P6KDOWLV$5HLVQHU+$VDGD³$+\GURVWDWLF3UHVVXUH$SSURDFKWR
&XIIOHVV%ORRG3UHVVXUH0RQLWRULQJ´IEEE EMBC, pp. 2173 ± 2176,
2004.
[17] 7 +HOGW ³&RPSXWDWLRQDO0RGHOV RI&DUGLRYDVFXODU5HVSRQVHWR
2UWKRVWDWLF6WUHVV´3K' GLVVHUWDWLRQ +DUYDUG-MIT Division of
Health Sciences and Technology, Cambridge, MA, 2004.
[18] 5 : 6WDIIRUG: 6+DUULVDQG$ 0:HLVVOHU ³/HIW9HQWULFXODU
Systolic Time Intervals as Indices of Postural Circulatory Stress in
0DQ´Circulation, vol. 41, pp. 485-492, 1970.
[19] L. S. Costanzo, Physiology, Elsevier, 4th Edition, 2010.
[20] T. Heldt, M. B. Oefinger, M. Hoshiyama and R. G. Mark,
³&LUFXODWRU\5HVSRQVHWR3DVVLYHDQG$FWLYH&KDQJHV LQ3RVWXUH´
IEEE Computers in Cardiology, vol. 30, no. 263, pp. 263-266, 2003.
2727