Wheelchair Exercise Monitor Development Platform

An Application for Wireless EMG Sensors

Amit Pal

, Kevin Monsalvo

, James Sunthonlap

, Paolo Arguelles

, Aldo Adame

, Jackson Tu

Ellie Tjara

, James Velasco

, Terrence Sarmiento

, Roxanna Pebdani

, Christine Dy

Stefan Keslacy

, Ray de Leon

and Deborah Won

Department of Electrical and Computer Engineering, California State University, Los Angeles, CA, U.S.A.

Department of Computer Science, California State University, Los Angeles, CA, U.S.A.

Department of Special Education, California State University, Los Angeles, CA, U.S.A.

Department of Kinesiology, California State University, Los Angeles, CA, U.S.A.

Keywords:

EMG, Wireless Sensor, Exercise Monitor, Individuals Who Use Wheelchairs, IMU, Heart Rate Monitor,

Wearable Sensors, Sensor Integration, Fitness Metric, Fitness Tracking Mobile App.

Abstract:

We present here a novel application for wireless EMG sensors. To combat the physical inactivity which has

tended toward cardiovascular disease in individuals who use wheelchairs, we have developed a monitoring

system to encourage these individuals to exercise. Wireless sensors are used to monitor kinematic or physi-

ological metrics, which inform the user of their activity levels during exercise and to track progress of their

ﬁtness levels over time. In particular, a new completely wireless, wearable EMG sensor (Dynoﬁt, Inc., TX) is

integrated with accelerometer and heart rate sensor data to monitor energy expenditure. The sensors communi-

cate with a custom designed mobile app which facilitates exercise at home, with the aim of helping individuals

who use a wheelchair to overcome what are commonly hindrances to exercising.

1 CARDIOVASCULAR FITNESS

IN WHEELCHAIR USERS

Individuals who use wheelchairs have an increased

risk of cardiovascular disease (Selassie et al., 2013;

Garshick, 2005). To address this growing health is-

sue, and to support preventative measures for car-

diovascular co-morbidity, we are developing a sys-

tem which would promote and facilitate exercise for

wheelchair users. (Abel et al., 2008; Blair, 1999).

A widely accepted recommendation for reducing car-

diovascular disease risk has been to increase average

daily energy expenditure by 300-350 kilocalories (RS

et al., 1993).

Thus, it has been important to the research com-

munity to determine what forms and amounts of ex-

ercise should be prescribed to reach this ﬁtness goal.

However, those who use wheelchairs face many barri-

ers to exercising the right way and right amount, many

tied to the challenge of getting to specialized gyms or

wellness centers.

To encourage and support wheelchair-dependent

persons in meeting these recommendations for caloric

expenditure, we are developing an in-home exercise

program, so that their workouts do not depend on

having access to expensive and/or large equipment,

and/or to therapists with whom scheduling or trans-

portation also disincentivizes exercising. In keeping

with the current trend of ﬁtness trackers, we are de-

veloping a system to track a measure of ﬁtness to mo-

tivate individuals to exercise and be encouraged by

progress in ﬁtness and/or activity levels. As energy

expenditure has been agreed upon as the most infor-

mative metric of cardiovascular ﬁtness (Abel et al.,

2003; Ainsworth et al., 1993), at the crux of this

in-home exercise monitoring system is the ability to

monitor energy expenditure continuously at home,

during exercise.

2 MEASURING ENERGY

CONSUMPTION

Currently available or researched activity monitors

and ﬁtness trackers for mobility impaired individuals

predominantly rely on acceleration, heart rate, or a

Pal, A., Monsalvo, K., Sunthonlap, J., Arguelles, P., Adame, A., Tu, J., Tjara, E., Velasco, J., Sarmiento, T., Pebdani, R., Dy, C., Keslacy, S., Leon, R. and Won, D.

Wheelchair Exercise Monitor Development Platform - An Application for Wireless EMG Sensors.

DOI: 10.5220/0006610000670073

In Proceedings of the 7th International Conference on Sensor Networks (SENSORNETS 2018), pages 67-73

ISBN: 978-989-758-284-4

combination of the two. Clinically relevant outcomes

for wheelchair users, such as amount of movement,

distance travelled, strength of maximum voluntary

contractions, and wheelchair propulsion have been ef-

fectively quantiﬁed with various sensing mechanisms,

the most common being the accelerometer or a reed

switch-based data logger(1). Accelerometry and the

data logger provide a reasonable measure of move-

ment and wheelchair propulsion (2), but motion de-

tectors are prone to false positives, as in the case of

an accelerometer-based step counter, which would de-

tect shaking of the device up and down and mistake

such motion as exercise. EMG is also better suited to

tracking compliance with prescribed exercises, since

the pattern of activation across multiple muscles can

be monitored. Furthermore, in contrast to accelerom-

eters, EMG may not only be used to monitor the con-

tractions of individual muscles, but also to capture

muscle activations during isometric contractions.

Energy expenditure is consistently relied upon

to measure and predict cardiovascular disease risk

(Sawka et al., 1980). However, measuring energy

expenditure directly through whole body calorime-

try or through oxygen uptake (VO

) measurements

requires expensive and/or impractical equipment and

facilities. Heart rate can be used to fairly accurately

compute energy expenditure (?). However, heart rate

is difﬁcult to obtain accurately during exercise, and

in particular, for spinal cord injury patients. Auto-

nomic dysfunction is common in spinal cord injured

patients (Krassioukov et al., 2008), and we have ob-

served anomolous heart rate recordings in spinal cord

injury subjects during exercise.

3 DREAM EXERCISE

MONITORING SYSTEM

DESIGN

The DREAM (Disability, Rehabilitation, and Engi-

neering Access for Minorities) exercise monitoring

system is designed to provide feedback which moti-

vates the user to improve his/her cardiovascular ﬁt-

ness through exercise. The main system components

are shown in Fig. 1.

Muscle activation, heart rate, and endpoint accel-

eration are measured in real-time from wireless sen-

sors, which include Flexdot EMG sensors (Dynoﬁt,

Carrollton, TX), a Wahoo Tikr heart rate moni-

tor (Wahoo Fitness, Atlanta, GA), and a custom-

packaged inertial monitoring unit. The DREAM app

allows users to set HR and activity target zones, con-

tinuously monitor their activity levels, and track the

Figure 1: System components of the DREAM exercise

monitor: A) Dynoﬁt’s Flexdot wireless EMG sensor; B)

the DREAM mobile app; C) the custom packaged wireless

IMU board; D) Mio Global’s Alpha 2 heart rate monitor

(Mio Global, Vancouver, British Columbia).

Table 1: Performance speciﬁcations.

EMG HR IMU

output

metric

integrated

envelope

heart rate wrist ac-

celeration

units % of max.

isometric

beats per

minute

m/s

sampling

rate (sam-

ples/sec)

60 1 4

input

range

±300µV 0-200bpm ±8g

resolution 440µV

(12-bit)

1bpm 940µg

(14-bit)

current

con-

sumption

(active

mode)

3mA <1mA 12.5mA

battery

life (hrs of

active use)

73hrs 300hrs 41 hrs

monitored metrics over time. The performance speci-

ﬁcations for the DREAM monitoring device are given

in Table 1

The ﬁtness-relevant physiolgical metrics are mon-

itored and wirelessly transmitted to a smart phone app

which depicts the metrics in an easy-to-read and mo-

tivating way. Fig. 3 illustrates the screen users will

see when exercising.

This cloud-based multi-user functionality and web

portal have yet to be implemented.

Here we describe the unique sensing components

of our design in greater detail.

SENSORNETS 2018 - 7th International Conference on Sensor Networks

Figure 2: Schematic system diagram of a next generation

DREAM exercise monitor.

3.1 EMG Sensors

EMG is acquired by commercial sensors developed

by Dynoﬁt to allow for completely wireless EMG ac-

quisition tailored for mobile applications. Other ven-

dors manufacture wireless EMG sensors, but these

sensors are more well suited for research applications,

as they require proprietary hardware (a base station)

and software. In contrast, the Dynoﬁt Flexdots com-

municate via Bluetooth Low Energy (BLE), and all

the hardware needed to perform wireless communica-

tion with a smartphone is contained within the wear-

able sensing unit. The sole requirement to acquire

data from the Flexdot is the capability of the receiver

to communicate via BLE.

The Flexdot module is fully packaged into a plas-

tic housing with snap female connectors for the elec-

trodes. Disposable electrolyte gel-ﬁlled electrodes are

snapped into the electrode contact terminals. Inside

the plastic housing is a custom board which provides

signal conditioning and A/D conversion, a BLE trans-

mitter, and a lithium coin cell battery.

3.2 Wearable IMU Module

During this development phase of the wheelchair ex-

ercise monitoring system, in order to determine the

most suitable ﬁtness metric, commonly used metrics

of activity will be acquired and compared to novel

metrics which incorporate or even rely centrally on

EMG measurements. Activity monitors most com-

monly measure heart rate and acceleration. An off-

the-shelf popular heart rate monitor, the Wahoo Tikr

is able to be used in the DREAM development plat-

form because the Tikr transmits data via the ANT+

wireless protocol. Thus, the DREAM app can com-

municate directly with the Tikr without need for any

Figure 3: Screenshots of two of the main DREAM app

screens: 1) activity metrics page displayed in real-time dur-

ing an exercise session; 2) the “Leaderboard” showing the

ranking of DREAM app users, ranked according to a ﬁtness

metric; 3) a log of stored data from past exercise sessions,

all accessible from the DREAM cloud server.

proprietary information. However, the top-of-the-line

accelerometers that are most often used in activity

monitoring in the literature pose the same issues that

the wireless EMG did; namely, they require connec-

tion to a PC and proprietary software. From pre-

liminary data, we had already seen that accelerom-

etry alone would not provide sufﬁciently accurate

feedback about energy consumption. Therefore, we

custom built a wearable BLE-based wireless inertial

monitoring unit module.

The system components and IMU module design

are provided in the board schematic in Fig. 5. This

custom PCB board consists of an IMU IC, BLE-

Wheelchair Exercise Monitor Development Platform - An Application for Wireless EMG Sensors

Figure 4: Flexdot components: the plastic housing for the

signal processing circuitry, with 3 female snap connections,

a 3V-coin cell battery, and the disposable snap electrodes

(Medtronic Covidien PLC, Minneapolis, MN).

Figure 5: Circuit schematic of IMU module.

UART wireless adapter, and Arduino microcontroller

board. The BNO055 is an intelligent 9-axis orienta-

tion MEMS sensor with a UART interface. We use

only the accelerometry measurements. In order for

our mobile app to acquire acceleration data wirelessly

from the BNO055, we interface the BNO055 with

a BLE transmitter on board the Bluefruit LE-UART

Friend (Adafruit Industries). The Bluefruit board is

a UART wireless adapter that establishes serial com-

munication with the BNO055 and then transmits this

data through the BLE transmitter to the mobile smart

phone. The ATMega 328P microcontroller, on board

an Arduino Pro Mini, controls the UART communi-

cation between the BNO055 and Bluefruit board.

This IMU module is powered by two 3.7V Li-ion

Polymer batteries in series, and ﬁts in a custom hous-

ing unit that can be worn via velcro strap around the

wrist. The Solid Works CAD drawing for the housing

unit is provided in Fig. 6. The package has slots for

the velcro strap and space for a power switch. The

housing is approximately 4cm x 4cm x 2cm.

Figure 6: CAD design for the IMU module housing unit,

3d-printed in ABS plastic. Dimensions are provided in mm.

Magnitude of 3D acceleration recorded during

wheelchair pushes, tricep extensions, and lat rows

for approximately 30 second bouts, with 30-second

rest intervals, is shown in Fig. 7. The DREAM IMU

module appears to record similar magnitude as Acti-

graph, but the Actigraph shows less high frequency

noise. We plan to implement low-pass ﬁltering in the

DREAM IMU module to obtain acceleration wave-

forms which even more closely match Actigraph’s.

Figure 7: Endpoint acceleration magnitude acquired by

Actigraph sensor vs. DREAM IMU module.

4 PERFORMANCE OF EMG

SENSOR

Before Dynoﬁt’s existence, a number of commercial

wireless EMG sensors existed on the market. These

are generally high performance sensors with excel-

lent noise cancellation, motion artifact suppression,

and overall signal to noise ratio. However, to the au-

thors’ knowledge, all of these wireless EMG sensors

require a receiver base station and a PC workstation

with proprietary software for data acquisition. We se-

lected the Dynoﬁt Flexdot for EMG sensing because

of its standalone capability; i.e., the Flexdot requires

no base station but transmits data wirelessly through

SENSORNETS 2018 - 7th International Conference on Sensor Networks

Bluetooth Low Energy (BLE) without requiring any

proprietary software. These features make it well

suited for our in-home exercise application, whereas

the other high performance EMG systems are better

suited for research applications or patient assessment

in a clinical setting but not practical for our applica-

tion of in-home exercise monitoring.

While the Dynoﬁt Flexdot sensors were more

practical and were the only sufﬁciently practical EMG

wireless sensors of which we were aware, we also

wanted to test the performance of the EMG sensors

relative to one of the top-of-the-line wireless EMG

commercial systems. We selected the Delsys Trigno

to which to compare the Flexdot.

A Flexdot was adhered to the muscle belly of the

right bicep; a Trigno sensor was adhered to the muscle

belly as well, adjacent to the Flexdot. EMG was ac-

quired from both systems for 60 seconds during bicep

curl exercises and isometric contractions. The EMG

envelope was obtained by full-wave rectifying the raw

EMG, and then applying a moving average ﬁlter with

a rectangular window of 100ms. Figure 8 illustrates

that the envelope was appropriately obtained from the

raw EMG. Both envelope amplitudes were normal-

ized to range between 0 and 1.

Figure 8: EMG envelope superimposed on the raw EMG

acquired from the Trigno sensor.

The envelope obtained from each of the sensors

were compared, as shown in Fig. 4. While the Flexdot

captured each muscle activation and showed temporal

accuracy, the amplitude of muscle activation some-

times exceeded that of the Delsys Trigno in this ap-

plication. Factors contributing to the amplitude dif-

ferences may include the differences in the location of

the sensors on a single muscle, spacing of electrodes

on the devices, size and the conductivity of the pads

used to adhere the device to the skin above the mus-

cle. These factors were not likely to explain the dif-

ferences, however, given that these amplitude changes

were observed for a given subject within the same

recording session. What appeared to be more likely

the case is that adaptive ﬁltering is being applied to

maximize use of the dynamic range on the amplitude

scale, such that the normalization of relative ampli-

tude is adjusted over time. Switching to non-adaptive

normalization is simply a matter of adjusting the post-

processing in ﬁrmware.

In order to quantify a direct comparison between

the Flexdot-based and Trigno-based EMG activity,

the Flexdot data was ﬁrst upsampled, since the for-

mer was acquired at 64Hz , and the latter at 2000Hz.

We then computed the RMS difference between the

two normalized envelopes, and obtained an RMS er-

ror of 0.27. Since the amplitudes are normalized, we

represent the RMS difference as 27% of the amplitude

range. As indicated above, this difference can be at-

tributed to the post-processing methods implemented

in ﬁrmware.

Figure 9: Comparison of EMG envelope acquired from the

Dynoﬁt Flexdot by the DREAM app and the EMG envelope

obtained from Delsys Trigno.

To help conﬁrm that the differences were more

likely due to differences in post-processing schema

than to physical characteristics, such as size and lo-

cation of the electrodes, we conducted another test.

This time, the subject performed bicep curls with elas-

tic arm bands (TheraBand, Akron, OH) at 3 levels of

increasing resistance. Recordings were taken from a

Trigno sensor placed on the left arm slightly proximal

to the center of the muscle belly, and a Flexdot sensor

placed just distal to the Trigno sensor, such that they

both overlapped with the center of the muscle belly.

On the right arm, we had the converse placement of

sensors, as seen in Fig. 10.

Figure 10: Placement of wireless EMG sensors. Left arm:

Trigno sensor placed more proximally, Flexdot sensor more

distally. Right arm: Flexdot sensor placed more proximally,

Trigno more distally.

Five bicep curls were conducted at each resistance

level. To increase the resistance level, we merely

shortened the theraband to ﬁxed lengths (of 80, 60,

and 40 cm). The resulting EMG envelopes are shown

in Fig. 11 for the left arm, and Fig. 12 for the right

arm.

Wheelchair Exercise Monitor Development Platform - An Application for Wireless EMG Sensors

Figure 11: Comparison of Trigno-derived vs. Flexdot-

derived EMG envelopes (from left arm) with Trigno sensor

more proximal, Flexdot more distal.

Figure 12: Comparison of Trigno-derived vs. Flexdot-

derived EMG envelopes (from right arm) with Trigno sen-

sor more distal, Flexdot more proximal.

The increasing amplitude of the Trigno envelope

corresponds with the increasing resistance of the arm

bands. It is possible that the Flexdot envelope has an

adaptive gain that was designed to maximize use of

the dynamic range at all times. The Flexdot performs

very well in terms of temporal resolution, which in

our application is critical. The location of the sensor

does have some inﬂuence on the EMG amplitudes,

but the main cause of the difference in amplitudes ap-

pears to lie within the post-processing in the sensors’

ﬁrmware.

How accurate the amplitude needs to be in order to

provide a motivating, reliable ﬁtness metric to poten-

tial DREAM app users has yet to be researched. Ad-

justment of the EMG envelope amplitude is expected

to require a straightforward adjustment of low-pass

ﬁltering parameters applied to compute the envelope.

5 CONCLUSIONS

Wireless sensing capabilities of EMG, heart rate,

and accelerometry have been integrated into a single

mobile app-based exercise monitoring system. The

DREAM system is being designed to help motivate

individuals who use wheelchairs to improve their car-

diovascular ﬁtness through exercise. We have pre-

sented the design and implementation of a ﬁrst proto-

type which will enable us to research the most appro-

priate ﬁtness metric on which to provide motivating

feedback to the users. In particular, we highlight the

selection and use of a new standalone wireless EMG

sensor which performs with signal quality compara-

ble to high-end wireless EMG sensors on the market,

with the added beneﬁts of low cost and practicality

in an in-home exercise application that is predicted to

help advance rehabilitation therapy.

ACKNOWLEDGEMENTS

This work was funded by an NIDILRR Field Initiated

Projects Program (Grant #90IFST0001-01-00). The

authors would like to gratefully acknowledge Dynoﬁt

founders and Flexdot developers Maria Schneider,

Edward Rosten, Alex Macdonell, Rohan Loveland,

and Mary Cooley who have generously offered their

time to help us learn more about their product.

We would also like to gratefully acknowledge Joel

Ramirez, Lloyd Ruiz, Lisa Le, and Isali Win who

have offered their time and effort to help in various

ways to carry out testing.

REFERENCES

Abel, T., Kr

oner, M., Rojas, V. S., Peters, C., Klose, C., and

Platen, P. (2003). Energy expenditure in wheelchair

racing and handbiking-a basis for prevention of car-

diovascular diseases in those with disabilities. Euro-

pean Journal of Cardiovascular Prevention & Reha-

bilitation, 10(5):371–376.

Abel, T., Platen, P., Vega, S. R., Schneider, S., and Str

uder,

H. (2008). Energy expenditure in ball games for

wheelchair users. Spinal Cord, 46(12):785.

Ainsworth, B., Haskell, W., Leon, A., Jacobs, D. J., Mon-

toye, H., and Sallis, J. (1993). Compendium of phys-

ical activities: classiﬁcation of energy costs of. Med

Sci Sports Exerc, 25:71–80.

Blair, S. (1999). Effects of physical inactivity and obesity

on morbidity and mortality: current evidence and re-

search issues. Medicine and science in sports and ex-

ercise, 31(11 suppl):S646–S662.

SENSORNETS 2018 - 7th International Conference on Sensor Networks

Garshick, E. e. a. (2005). A prospective assessment of

mortality in chronic spinal cord injury. Spinal Cord,

43:408–416.

Krassioukov, A., Karlsson, A.-K., Wecht, J. M., Wuermser,

L.-A., Mathias, C. J., and Marino, R. J. (2008). As-

sessment of autonomic dysfunction following spinal

cord injury: Rationale for additions to international

standards for neurological assessment. Journal of

Rehabilitation Research & Development, 44(1):103–

112.

RS, P. J., RT, H., AL, W., IM, L., DL, J., and JB, K.

(1993). The association of changes in physical-

activity level and otherlifestyle characteristics with

mortality among men. N Engl J Med, 328:538–545.

Sawka, M. N., Glaser, R. M., Wilde, S. W., and von Luhrte,

T. C. (1980). Metabolic and circulatory responses to

wheelchair and arm crank exercise. Journal of Applied

Physiology, 49(5):784–788.

Selassie, A., Snipe, L., Focht, K. L., and Welldaregay, W.

(2013). Baseline prevalence of heart diseases, hyper-

tension, diabetes, and obesity in persons with acute

traumatic spinal cord injury: potential threats in the

recovery trajectory. Top. Spinal Cord Inj. Rehabil.,

19:172–182.

Wheelchair Exercise Monitor Development Platform - An Application for Wireless EMG Sensors