Glabella Continuously Sensing Blood Pressure Behavior using an Unobtrusive Wearable Device Christian Holz Edward Wang Microsoft Research
Glabella
continuously records the pulse at 3 locations, pulse transit time (PTT) continuously predicts blood pressure behavior (inversely correlated with PTT)
example: getting up after leaning back optical signals 2.8 2.0 1.2 0.4 computed pulse transit times 38 34 30 26 5 sec 10 sec 15 sec 20 sec
example: getting up after leaning back optical signals 2.8 2.0 1.2 0.4 computed pulse transit times 38 34 30 26 5 sec 10 sec 15 sec 20 sec getting up
example: getting up after leaning back optical signals 2.8 2.0 1.2 0.4 computed pulse transit times 38 34 30 26 5 sec 10 sec 15 sec 20 sec getting up blood pressure restores
introduction
blood pressure ~75 million American adults have high blood pressure (32%) high blood pressure was a primary cause of death for more than 410,000 Americans in 2014 1 in 5 American adults is unaware of having high blood pressure https://www.cdc.gov/dhdsp/data_statistics/fact_sheets/fs_bloodpressure.htm
blood pressure capture invasive gold standard (clinical settings) semi-automated cuff-based monitor
blood pressure during the day https://www.health.harvard.edu/heart-health/experts-call-for-home-blood-pressure-monitoring
blood pressure during the day additionally influenced by short-term events eating and drinking Smith et al., Physiology & Behavior 1997 events at work Lindquist et al., Hypertension 1997 exercise or postural changes Imholz et al., Cardiovascular Research 1990 medication intake
continuous monitoring is desirable
readily available and accurate suitable for long-term tracking needs correct attachment takes time less suitable in a mobile setting oscillometric cuff device
pulse transit time := the time it takes a pulse wave to travel between two sites of the body t 0 requires no cuff and can be measured during each beat t 1 repeatedly shown to correlate highly with blood pressure the hope
pulse transit time vs. blood pressure PTT vs. SBP 280 r = - 0.670 0.766 TD ( ms ) 260 240 220 200 E - T PTT Nitzan et al. Physiol Meas 2002 Shahrbabaki et al. EMBC 2016 Mukkamala et al. Trans Biomed Eng 2015
McCombie et al. EMBC 2006 Liu et al. Sensors 2015 Beckmann et al. EMBC 2017
integrated form factor requires no user interaction
superficial temporal artery occipital artery
superficial temporal artery occipital artery angular artery
Glabella cuff-less, wearable, continuous blood pressure monitoring
dominant goal: a socially acceptable form factor a lightweight and unobtrusive form factor that directly integrates into the frame of glasses without causing distraction to either the wearer or surrounding people lasts a full day on a single charge while collecting continuous measurements goal
V1 V2
V2 V3
V3 V4
fully integrated & standalone Glabella V5 prototype
Glabella prototype main board and battery optical sensors
Glabella prototype continuously collects 3 optical reflections (at up to 5 khz) 3-DOF inertial motions (at 200 Hz) 15 hours runtime 45 grams in weight
electronics
Glabella s main board power switch FPC connectors SD IMU PSoC USB
Glabella s main board 305 mah LiPo battery real-time clock
Glabella s optical pulse sensor board skin-facing back photodiode LED opamp & filter circuit
Glabella s optical pulse sensor board flex PCB cables connect to the main board
mechanical design
off-the-shelf metal-frame glasses parametrized frame 3D printed with digital ABS mechanical design
to fit a wearer s head dimensions to ensure sensor contact mechanical design parameters
testing amplification circuit in-lab pilot tests quick dimensioning adjustable prototype frame
underfilled sensor underfilled hot glue to sustain constant skin contact
3D printed nose pad with embedded, underfilled optical sensor
signal processing
3 optical signals angular artery (nose pad) superficial temporal artery occipital artery (behind ear) s ang s sta s occ
1filter optical signals 2analyze the dominant frequency of each optical signal 3 4 5 detect pulses and extract temporal features validate candidate features compute pulse transit times signal processing
raw signal recordings filtered signals voltage (0..3.3V) 1.0 0.9 0.8 0.7 0.6 0.5 0.4 1.9 1.8 1.7 1.6 1.5 1.4 1.3 11 sec 2 sec 3 sec 4 sec 5 sec filter optical signals 1 sec 2 sec 3 sec 4 sec 5 sec bandpass filter (0.4 Hz, 8 Hz)
apply Fast Fourier transform to each 15-second window to extract dominant frequencies f ang, f sta, f occ derive heart rate if f ang f sta < ε and f ang f occ < ε otherwise discard this window for feature extraction 2derive the dominant frequencies
0.6 0.4 0.2 0 0.2 0.4 0.6 3 0.2 sec 0.4 sec 0.6 sec 0.8 sec detect pulses, extract temporal features
0.6 0.4 0.2 0 0.2 0.4 local peak 0.6 2 nd derivative peak t ang 3 0.2 sec 0.4 sec 0.6 sec 0.8 sec detect pulses, extract temporal features
discard if IMU variance exceeds a threshold correlate normalized optical signals to assess signal quality phase offset 50 highest-correlating phase shifts 4validate candidate features
pulse transit time between the sensors on superficial temporal artery & angular artery occipital artery & angular artery PTT ang sta := t ang t sta PTT ang occ := t ang t occ 5compute pulse transit times
example: getting up after leaning back bandpass filtered signals 2.8 2.0 1.2 0.4 computed pulse transit times 38 34 30 26 5 sec 10 sec 15 sec 20 sec getting up blood pressure restores
in-the-wild evaluation
sustain operation and capture useful data during everyday wear and regular activities determine the correlation between the pulse transit times recorded by our prototype and systolic blood pressure values measured by a cuff-based monitor goals
wear the custom-fit prototype glasses at least 12 hours per day record blood pressure values three times an hour at least 30 measurements per day 5 days of participation task
Glabella device oscillometric device task
screening session for custom fitting and instructions procedure
charger & cables blood pressure cuff custom-fit Glabella device procedure
pick up the box wear the custom-fit prototype and record measurements from Monday through Friday procedure
4 participants 2 female (ages 25 and 39) 2 male (ages 40 and 42) all had Fitzpatrick Skin Type II no known related medical conditions (e.g., hypertension) Microsoft employees in different capacities $400 gratuity depending on compliance participants
analyzed surrounding ±2 minutes of each blood pressure measurement to predict the wearer s heart rate systolic blood pressure results
heart-rate correlations
heart-rate correlations
heart-rate correlations
heart-rate correlations
heart-rate correlations
histogram: systolic blood pressure
PTT ang sta vs. systolic blood pressure
PTT ang sta vs. systolic blood pressure
PTT ang sta vs. systolic blood pressure
PTT ang occ vs. systolic blood pressure
PTT ang occ vs. systolic blood pressure
conclusion
feasibility to continuously collect optical pulse reflections to track heart rates and extract meaningful differences in time-of-arrival established that Glabella tracks blood pressure behavior cuff-less, passively, and conveniently without demanding user input throughout the day and regular activities
numerous benefits sensing on the angular artery constant force due to gravity no adjustment necessary best PTT results sensing on angular artery and superficial temporal artery
for absolute blood pressure values, simple linear per-user model can predict systolic blood pressure values within ±10 mmhg (assuming correct cuff use)
number of participants controlled in-lab study with a continuous baseline calibration is a significant challenge limitations & future work
Glabella Continuously Sensing Blood Pressure Behavior using an Unobtrusive Wearable Device Christian Holz http://www.christianholz.net
Alexander Ching, Christopher O Dowd Mike Sinclair, Jason Goldstein, Sokunthea Neang Daniel Cletheroe, Pavel Kulik, Andrew Carek Kambria Tabor Acknowledgments
Glabella Continuously Sensing Blood Pressure Behavior using an Unobtrusive Wearable Device Christian Holz http://www.christianholz.net