Content uploaded by Mariusz Chmielewski
Author content
All content in this area was uploaded by Mariusz Chmielewski on Nov 08, 2017
Content may be subject to copyright.
The method for processing electromyography and inertial sensors
supporting chosen set of neurological symptoms for clinical trials support
and treatment assessment
MARIUSZ CHMIELEWSKI
MICHAŁ NOWOTARSKI
Cybernetics Faculty
Military University of Technology
gen. S. Kaliski Street 2, 01-476 Warsaw
POLAND
mchmielewski@wat.edu.pl
michal.nowotarski@wat.edu.pl
Abstract: This paper discusses method and tool for supporting and assisting clinical trials of pharmaceutical
drugs utilising wearables and mobile technologies. Biomedical sensors and handhelds deliver new opportunities
to gather and process medical data. Presented method implements such observations and delivers new,
convenient means for remote patient monitoring. Clinical trials require methodology and tools to support
iterative, accurate assessment of drug intake and treatment effects. Currently available methods rely mostly on
analogue approach (booklets) in which a trial participant is reporting symptoms in a booklet. Such approach
often can be biased by unpunctual, not precise reporting. Handhelds can support trial processes by
automatically scheduling and recording an actual time of reports and most of all it can record the inertial and
biometric sensor data during the survey process. Presented analytical method (tremors recognition) and its
software implementation offer consistent approach to clinical trials assistance transforming Android
smartphone into remote reporting and notification tool. PatronDroid tool supplements also features for sensor
based disease diagnostics identifying PD tremors as well as specific tonic or clonic phases. The tool’s feature
composition delivers convenient and reliable utility which can assist patients and medical staff during the
process objectifying the clinical tests, helping to ensure good quality of the clinical trial data, instantly
available and easily accessible.
Key words: mobile applications, wearable sensors, telemedicine, clinical trials, neurological symptoms
1 Introduction and main concept
The main idea presented in the work is to utilise
handhelds and wearable biometric and inertial
sensors in order to assist clinical tests and medical
examinations [1][7][8]. Such an approach can
deliver promising results in order to obtain objective
data during massively executed clinical tests. The
tool can also lower costs of conducting and
analysing results of clinical tests, as the logistics of
the tests can be simplified. Presented method,
algorithms and its implementation delivers handheld
application and a sensor armband, which are
managed by the central server service. The central
server service can be further integrated with any
SOA infrastructure delivered by clinical test
provider. Such construct is flexible easily
configurable and can be deployed as a cloud
services to support scalability.
The use of accelerometer and/or
electromyography in neurological diseases treatment
were mentioned in various publications. There were
discussed ideas to monitor Parkinson’s disease with
the smartphone [2] or to provide automatic detection
of tremors and dyskinesias with accelerometer and
electromyographic signal of specified monitoring
device [3]. Idea of use an accelerometer during
detection of epilepsy’s seizures were mentioned in
[4][5][6]. What is more, since 2012 Military
University of Technology has developed several
sensor based solutions for neurological disease
assistance and monitoring - SENSE [7], PATRON
WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE
E. Korobko, S.Vilanskaya, A. Baeshko, N. Shestak
E-ISSN: 2224-2902
134
Volume 14, 2017
[8] utilising smartphone’s inertial sensors
proprietary wireless biometric sensors.
2 Tool for assisting clinical tests
PATRON [8] (smartphone-based mobile system
utilising wearable sensors) offers innovative
approach of Parkinson’s Disease symptoms
examination delivered through application of
myography and inertial sensors integrated with
handheld devices. The utilisation of sensor-assisted
diagnostics of tremors and motor skills, delivers
new opportunities for neurological research
especially in the domain of effective drug usage and
clinical testing of pharmaceutical therapies. The
developed hardware, the PATRON multi-sensor is
first fully mobile, wireless sensor fusing several
channels of data gathered from surface
electromyography, mechanomyography, 3-axis
inertial sensor and magnetometer. The sensor can be
seamlessly attached to a forearm or calf in order to
remotely gather, filter and transfer the data to a
handheld. The multistage DSP analysis of the data is
performed in real time, delivering data of tremors
frequency and magnitude, the intensity of muscle
tensions and the orientation of the sensor. Fused
data describing the neurological examination is than
stored, aggregated with the time and location and
uploaded to the reporting server. The mobile
application delivers two main functionalities ad-hoc
tremor examination and on-demand clinical test
surveying. Therefore, the application can be used
for examining, the tremor intensity and limb
movement fluency, evaluating the Parkinson’s
Disease on-off fluctuations and dyskinesia. System
supports the PD patient by estimating current health
state and in consequence providing recommendation
for medicine usage. PATRON can also be used as a
clinical test support toolkit for researchers and
pharmaceutical companies guiding patients through
the complete process of reporting the drug
effectiveness. Sensor-based tests complement the
patient’s surveys, delivering objective information
about the patient’s health state therefore evaluating
the effectiveness of a drug and applied treatment.
The system offers also a server based reporting
services which serve as a clinical tests warehouse.
Web portal visualizes the clinical test process
organizes the data and evaluates the characteristics
of the medicine effectiveness [12][16]. PATRON
server has been deployed on Microsoft Azure Cloud
environment, which provides high-scalability and
security characteristics. Web application can be
accessed by the researchers, medical staff and
pharmaceutical companies in order to inspect and
monitor current state of clinical test process.
3 Utilization of various devices to
receive signals
The MYO armband is a gesture and movement
control device. Its purpose is to receive signals
during arm’s movement and gestures. The device
contains eight EMG sensors (device segments)
detecting muscle tension with sample rate 200 Hz
and accelerometer, gyroscope and magnetometer
inside built-in IMU with sample rate 50 Hz
(Invensense MPU-9150 9-dof motion sensor). It is
controlled by the Free-scale Kinetis Cortex M4
120MHz processor and communication is realized
with Micro USB port or Bluetooth LE protocol.
Armband can be expandable between 7.5 - 13
inches (19 - 34 cm) forearm circumference and
weights 93 grams [9][10].
The vendor has provided SDKs dedicated for
various mobile platforms and capabilities. For
example, the iOS SDK contains methods providing
ability to read data from EMG sensors, while the
Android SDK lacks such features. According to
[11], sensors of the Myo have the following
characteristics:
- electromyograph – unit of measurement: mV,
range: +/- 0.45 mV,
- accelerometer – unit of measurement: g, range: +/-
16 g,
- gyroscope – unit of measurement: °/s, range: +/-
2000 °/s,
- position sensor – unit of measurement: quaternion.
The manufacturer claims that sensor’s battery
should last a whole day of continuous use (which is
one of crucial requirements for the clinical test
assistance). For the band in a sleep mode,
disconnected from any devices, expected operating
time is about a week. There are two batteries inside,
with a capacity of 260 mAh each. The
communication range via Bluetooth is up to 15
meters. Armband’s sensors can work on the leg
what can be useful in various diagnostic purposes.
During Patron Droid’s diseases evaluation there
are also being used inertial sensors from
smartphones and smartwatches.
WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE
E. Korobko, S.Vilanskaya, A. Baeshko, N. Shestak
E-ISSN: 2224-2902
135
Volume 14, 2017
Fig. 1. Location of sensors on the patient’s arm
All sensors are placed at the appropriate
locations of the user's body. For the operation of the
algorithm, it is necessary to locate sensors in
planned places. Myo's headband is made to be worn
on the forearm. Smartwatch has to be located on the
wrist, while the smartphone must be held by the user
in his hand.
4 Monitoring of the patient’s health
The application allows to collect patient’s state data
in two ways:
- periodically, when a patient is asked once in a
defined time to complete a survey about his
condition
- forced, when the patient at his own request initiate
the test procedure.
Fig. 2. Patron Droid survey screen
Periodic mode means that a new survey is
displayed to the patient in a regular time intervals.
The survey contains of basic questions about the
patient's health. At the moment these questions are
written in Polish, because the research is currently
being conducted in Poland. The questionnaire at the
beginning contains questions that concern:
- a patient’s ability to complete the survey
- subjective assessment of the patient's condition
- evaluation of dyskinesias
These questions can be modified and adapted to
the needs of the patient. In addition, during the
completion of the survey by the patient,
simultaneous collection of sensor’s data is available
at the same time. These signals are being recorded
in the device and the correlation between the
subjective responses given by the patient and the
objective signals collected from the sensors is
checked by the clinical test process manager
[12][16].
This approach allows for a continuous
verification of the patient's state of health. On this
basis, it is able to assess the severity of the disease,
the effectiveness of drugs and monitor the disease
progresses what can provide useful information
about a treatment to choose.
In addition to the periodic mode, it is also
possible to perform a forced test. This means that
the patient first selects the configuration in which
the test will be performed (sensors to use, which
arm, one or both arms test, time of the test),
followed by signal collection at a predetermined
time.
Fig. 3. Patron Droid examination screen
WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE
E. Korobko, S.Vilanskaya, A. Baeshko, N. Shestak
E-ISSN: 2224-2902
136
Volume 14, 2017
During the test, the application provides basic
information such as the test time, current sensor
reading, and graphs showing the signal
characteristics throughout the test. If the patient has
previously selected the appropriate option, received
signal values can be stored in the device memory
during the test.
Apart from the possible data recording, the
signals received during the test are continuously
transmitted to the reasoner component. This is
where the signal analysis is done, what is described
in more detail in the next section.
During the examination, the algorithm
determines the individual parameters for specified
time frame. At the end of the operation time, the
reasoning is performed on the obtained parameters,
and the result of the test is displayed to the user.
Fig. 4. Patron Droid examination result screen
Features and views of the application was
described in details in [17].
4.1 The algorithm of disease detection
The purpose of the algorithm’s current version is
to detect symptoms of Parkinson's disease and
epilepsy. To aim the purpose, two types of signals
are used if they are possible to receive:
- inertial signals that tell where and how fast the
patient's limb was moving,
- electromyographic signals presenting limb’s
muscles tension.
Based on the information collected from these
signals and the correlation between them, a final
diagnosis is made. The possibility of receiving
selected results depends on the measuring device
that is used for the test. This is due to the fact that
Myo has both inertial and electromyographs sensors
when smartphones and smartwatches allow to
collect signals only from inertial sensors.
Having recorded both signals, it is possible to
distinguish some features of epilepsy and
Parkinson's disease:
- Parkinson's disease is characterized by slight
tremors without increased muscle tension.
- The epilepsy tonic phase is associated with a
strong muscle strain of the patient who remains
without any move.
- The clonic phase of epilepsy is associated with
substantial tremors of the patient's limbs as well as
rapid muscle strain.
Based on these features, the algorithm is trying
to make a diagnosis. Where the test is carried out
with a Myo armband, it is possible to check all of
them. When a smartphone or smartwatch is used for
a test, the diagnosis is based only on inertial signals,
which limits monitoring the patient's behaviour to
the characteristics of his limb movement.
WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE
E. Korobko, S.Vilanskaya, A. Baeshko, N. Shestak
E-ISSN: 2224-2902
137
Volume 14, 2017
Fig. 5. Flow of the decision algorithm
This means that depending on the test
configuration, it is possible to receive results with
an different accuracy.
Fig. 6. Possible decisions of the algorithm
In every test configuration, the examination is
always similar. In each time window are gathered
signals from the accelerometer and - if possible -
from the electromyograph. For each time window,
the features of the collected signals are determined
and stored in the device memory. At the end of the
examination, these features are appropriately
analysed and correlated between them, and the user
appropriate result is presented to the user. The
present form of the algorithm is described in more
detail in [17]. Obtained results depend on the
occurrence of the problems identified during the
examination:
- No abnormalities means there was not set any of
the examination flags.
- The Parkinson’s disease or the epilepsy suspicion
based on arm trembling during examination of one
arm without the armband sensor.
- The Parkinson disease suspicion implies that there
was arm trembling without muscle tension during
examination of one arm with the armband sensor
or only one arm trembling during any examination
of both arms.
- The epilepsy tonic phase suspicion is based on
increased muscle tension without arm trembling
during an examination with the armband sensor.
- The epilepsy clonic phase suspicion means that
there was variable muscle tension and arm
trembling during an examination with armband
sensor use.
Reduce a time of the
window function
Analyse if obtained
parameters
exceeded calibrated
threshold values
Receive 8 channel
electromyographic data
Correlate obtained
parameters for inertial
and electromyographic
signals
Identify an active
hand and recalibrate
algorithm
Receive inertial data for
the first hand
Evaluate quantitative
features of intertial data
for a window function
Evaluate quantitative
features of intertial data
for the window function
Evaluate quantitative
features of
electromyographic data
for the window function
Provide an
examination result
Evaluate amplitude,
average, min and max
values for the window
function
Evaluate amplitude,
averag e, min and max
values for the window
function
Choose sensors
[smartphone,
smartwatch, Myo] and
examination time
[average amplitude
< threshold value]
[selected sensor:
smartphone or
smartwatch]
[current time <
examination time]
[average amplitude
> threshold value]
[current time >
examination time]
[selected
sensor: Myo]
Abnormalities
not detected
Parkinson's
disease
suspicion
Epilepsy
clonic phase
suspicion
Epilepsy tonic
phase
suspicion
[tremors
detected]
[muscles
not tensed]
[muscles
not
tensed]
[detected
tensed
muscles]
[detected
tensed
muscles]
[tremors not
detected]
WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE
E. Korobko, S.Vilanskaya, A. Baeshko, N. Shestak
E-ISSN: 2224-2902
138
Volume 14, 2017
4.2 Further development of the algorithm
The current algorithm is not sufficient for advanced
diagnosis purposes, so a plan for its further
development has already been prepared. In the next
stage the algorithm will be significantly expanded.
The diagnosis of Parkinson's disease will use an
actigraphy to measure mobility of the limb. In the
case of epilepsy, the purpose is to detect particular
types of seizures. To achieve this, during the test
will be determined muscle activation function and
measured intervals between succeeding slopes on
the signal record. Based on these data and
accelerometer the tonic and clonic phases will be
detected. The algorithm will search for similar
characteristics as currently (described in 4.1).
Atonic seizures will be detected by sensors located
on the patient's neck, as a best place to observe their
characteristics. This is a sudden momentary stroke
of muscle tension along with the following drop on
the EMG signal. It will be correlated with patient's
behaviour (usually a fall-off) found on the inertial
sensor. The algorithm will also detect myoclonic
seizures that are characterized by sudden muscle
spasms in the electromyographic signal. It is also
intended to use the machine learning technique in
patterns comparing.
Considered decision support mechanism was a
decision tree. After the research, it was decided it
will not be used, because each decision have to be
explicitly described in advance. Computer systems
which supports of making a decision generally
require gathering and ongoing access to all
knowledge about the issues to be resolved.
Nowadays, work is being done on the use of neural
networks for pattern recognition. Neural networks
seem to be suitable for the algorithm due to their
ability to adapt dynamically to learning patterns. In
addition, they can be used when the data was not
present in the teaching pattern. They are able to
match the most similar pattern and make a decision.
Neural networks require a one-time learning, but
they show tolerance for discontinuities, accidental
disorders, or even lacks of a knowledge in the
learning set. In the case of neurological diseases, the
course of which may be individual for each patient,
it seems more appropriate to use neural networks.
The algorithm will be based on more signal
characteristics than the current version. Based on
this, there will be an inference related to disease
detection. In the Section 5 are presented selected
signal features:
- number of zero crossing,
- mean waveform length,
- mean absolute value,
- mean frequency,
- mean power,
- peak frequency,
calculated for test cases. These characteristics are
part of these which will be used during future
development of the algorithm.
At the moment, software authors do not have
large sets of research data yet. At this point the
research is done in several dozen attempts on
patients who reported to study.
In addition, research are still conducted to make
it possible to identify if obtained sensors data allows
to accurately identify conditions in patients’
behaviour which are needed to diagnose properly.
One of the issues is to determine the visibility of
Parkinson's tremors in the electromyography record.
Currently, the goal of the algorithm is to record
seizures to determine the effectiveness of the
treatment. Based on the recorded signals, a doctor is
able to arbitrarily determine whether the treatment
provided has produced the desired effect in an
objective way, not subjective from the patient's
perspective.
Attempts will be made to ensure that the
algorithm next to the seizure recording also detects
the symptoms of the impending attack. This means
that during the analysis, appropriate signal
characteristics will be detected to indicate that the
patient will have a seizure. He would be noticed and
be able to prepare himself for the seizure.
Evaluation of Parkinson’s disease progress will
be based on:
- Frequency and intensity of Parkinsonian tremors
(sensor-based)
- Occurrence of ON and OFF phases (time,
duration, intensity)
- Sleep cycle schedule – as evaluation of the sleep
quality
- Analysis of the drug effect on the occurrence of
dyskinesia
- Self-assessment survey analysis to rate the
patient’s comfort level
- Analysis of survey completion to evaluate the
patient’s compliance
The application is intended to be a patient's
helper during his life with the disease. In the event
that a seizure is registered, the application will
automatically send an information to persons
defined in the contact list. In addition, during a
seizure, through a device's speaker a message will
be issued informing observers of the attack what is
happening with the instructions for them how they
can help the patient.
WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE
E. Korobko, S.Vilanskaya, A. Baeshko, N. Shestak
E-ISSN: 2224-2902
139
Volume 14, 2017
5 Experiments and tests
Tests were made by the application developer and
was simulated to cover different examination cases.
Every test was made with different arm movement,
different muscle tension and the same test
configuration. Test were conducted with the
armband sensor use. Charts were made based on the
data received and saved from sensors during
examination.
5.1 Test case 1
The first test case was the recording of a healthy
person's behaviour. The arm was held stably,
without any movement and the accelerometer signal
(blue) was maintained throughout the examination
at a low level. Low values on the electromyograph
chart (orange) indicates that throughout the study
muscles were not significantly tightened. Similar
results have been obtained for each time window
sample. For the accelerometer the number of zero
crossings in all cases is close to 100.
Table 1. Statistics of the Test case 1 for ACC
Time window 1 2 3 4
No. of zero
crossings
9
4
97
97
85
Mean waveform
length [s]
0.54 0.51 0.51 0.59
Mean absolute
value [m/s2]
0.08 0.06 0.05 0.05
Maximum
amplitude [m/s2]
0.19 0.21 0.19 0.21
Median
frequency [Hz]
21.45 24.96 21.91 17.26
Mean frequency
[Hz]
21.01
22.20
23.72
18.83
Mean power
0.07
0.01
0.01
0.01
Peak frequency
[Hz]
19.27 19.51 19.70 17.09
Table 2. Statistics of the Test case 1 for EMG
Time window 1 2 3 4
No. of
zero
crossings
78
79
88
106
Mean waveform
length [s]
0.0
0.0
0.06
0.0
Mean absolute
value [mV]
0.0
0.0
0.0
0.01
Maximum
amplitude [mV]
0.0 0.0 0.0 0.0
Median
frequency [Hz]
20.51 19.46 21.0 23.1
Mean frequency
[Hz]
17.5 20.0 22.0 23.8
Mean power
0.00
0.00
0.00
0.00
Peak frequency 16.07 15.9 17.71 21.5
[Hz]
Even the healthy person’s arm involuntarily moves
in a small range and recorded vibration frequencies
are about 22 Hz. The average value of tremors
during the test was negligible, about 0.06 m/s2.
Maximum received amplitude exceeds 0.2 m/s2. As
is the case of the accelerometer signal, the results
obtained with the electromyograph also show a
negligible muscle tension during arm’s holding. The
number of zero crossings is about 90, and in all
cases, the average value for every time window is
minimal and its value is 0.01 mV. This shows that
during the test muscles were hardly strained.
Fig. 7. Graph of extracted 150 samples started at 1000th
sample of the Test case 1. Blue line is an accelerometer
signal, orange line is an electromyograph signal.
5.2 Test case 2
The simulation of a person suffering from the
Parkinson’s disease. During the test, arm was still
raised and entered in a gentle tremor. Muscles were
relaxed during whole examination. Compared to the
Test case 1, a reduced number of zero crossings can
be observed in the case of the accelerometer. This
means that tremors were slower, but based on mean
absolute value can be stated that it was stronger than
in a healthy person examination.
Table 3. Statistics of the Test case 2 for ACC
Time window 1 2 3 4
No. of zero
crossings
64 67 67 67
Mean waveform
length [s]
0.08 0.07 0.07 0.0
Mean absolute
value [m/s2]
0.37 0.8 0.7 0.7
Maximum
amplitude [m/s2]
1.7 1.9 2.1 2.32
Median
frequency [Hz]
15.8
14.8
14.8
14.45
-0.04
-0.03
-0.02
-0.01
0
0.01
0.02
0.03
0.04
20:112 20:268 20:475 20:685
-0.15
-0.1
-0.05
0
0.05
0.1
0.15
Tremor [m/s2]
Time [s]
Voltage [mV]
ACC EMG
WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE
E. Korobko, S.Vilanskaya, A. Baeshko, N. Shestak
E-ISSN: 2224-2902
140
Volume 14, 2017
Mean frequency
[Hz]
14.2 14.90 14.7 14.4
Mean power 0.37 0.81 0.72 0.73
Peak frequency
[Hz]
13.53 13.47 13.6 13.4
Table 4. Statistics of the Test case 2 for EMG
Time window 1 2 3 4
No. of zero
crossings
18 22 26 19
Mean waveform
length [s]
0.27 0.22 0.20 0.2
Mean absolute
value [mV]
0.02 0.01 0.0 0.0
Maximum
amplitude [mV]
0.05
0.05
0.06
0.05
Median
[Hz]frequency
5.87
5.59
7.53
7.17
Mean frequency
[Hz]
5.90
5.55
7.29
6.62
Mean power 0.00 0.00 0.00 0.00
Peak frequency
[Hz]
3.82
4.44
5.27
3.8
Fig. 8. Graph of extracted 150 samples started at 1000th
sample of the Test case 2. Blue line is an accelerometer
signal, orange line is an electromyograph signal.
The maximum reached amplitude approached 2
m/s2. On the basis of frequency measurements, it is
possible to confirm that author failed to precisely
reproduce the behaviour of the patient, as values
around 14 Hz are higher than model values for the
Parkinson's disease. Similar results as with the
inertial signal occurred with the electromyograph.
The signal is slightly higher than in the case of a
healthy person test (mean 0.02 mV and maximum
amplitude 0.05 mV). But shivering was slower as
the number of passes was zero.
5.3 Test case 3
The test was an example of algorithm behaviour in
case of the epileptic tonic-clonic seizure. The
seizure consists of two parts and its picture every
time looks similar: it starts with the tonic phase and
then passes into the clonic phase. The duration of
each phase is not determined, it is an individual
feature of the patient. In the test case, the data
obtained during the test were divided into two
separate parts: for the tonic and the clonic phase, to
distinguish behavioural differences of different
stages of the seizure. In the initial phase of the
attack, the muscle tension significantly exceeds the
threshold values and the corresponding flag is set to
true. However, there is no arm movement at the
same time, so that no tremors are detected. Only in
the second, clonic phase, tremors are present and
they are rapid along with significant strains and
muscle relaxation. Both these features are detected
in the phase and flags are set. For the first stage
application detected an tonic phase of the seizure
and a clonic phase for the second.
5.3.1 Tonic phase
The simulation of a person in the tonic phase of
epilepsy. Arm was held in one position. To keep
muscles tightened, a fist was clenched. Muscles
were tensed all the time with a full strength. After
muscles having tensed, this state was maintained for
about 8 seconds and then muscles were relaxed.
Accelerometer measurements for the test phase
shows that despite extensive muscle tension, slight
arm tremors occurred during the examination. Mean
absolute value and maximum amplitude were higher
than in Test 1 but tremors were shorter, what can be
observed by a greater number of zero crossings. The
most important feature of the test is observable in
the case of the electromyographic signal. The
number of zero crossing is very low for the test,
which means that the signal value has remained on
high values during whole examination. For the
second time window, it never dropped to zero.
Signal values were significantly higher than for
previous test cases.
Table 5. Statistics of the Test case 3 for ACC
Time window 1 2 3 4
No. of zero
crossings
79 116 112 34
Mean waveform
length [s]
0.06 0.04 0.05 0.0
Mean absolute
value [m/s2]
0.18
0.16
0.16
0.07
Maximum
amplitude [m/s2]
0.99
0.56
0.77
0.28
Median
15.91
24.67
24.64
14.76
-0.04
-0.03
-0.02
-0.01
0
0.01
0.02
0.03
0.04
17:014 17:220 17:384 17:606
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
Tremor [m/s2]
Time [s]
Voltage [mV]
ACC EMG
WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE
E. Korobko, S.Vilanskaya, A. Baeshko, N. Shestak
E-ISSN: 2224-2902
141
Volume 14, 2017
frequency [Hz]
Mean frequency
[Hz]
16.97 25.03 24.07 17.47
Mean power
0.55
0.33
0.37
0.29
Peak frequency
[Hz]
15.88 23.33 22.52 13.16
Table 6. Statistics of the Test case 3 for EMG
Time window 1 2 3 4
No. of zero
crossings
10 0 14 22
Mean waveform
length [s]
0.26
5
0.15
0.11
Mean absolute
value [mV]
0.08
0.11
0.06
0.01
Maximum
amplitude [mV]
0.29
0.27
0.20
0.03
Median
frequency [Hz]
0 0 0 13.45
Mean frequency
[Hz]
7.67 0 3.19 11.81
Mean power 0.01 0.02 0.01 0.00
Peak frequency
[Hz]
2.02 0.20 2.84 8.63
Fig. 9. Graph of extracted 150 samples started at 1000th
sample of the Test case 3. Blue line is an accelerometer
signal, orange line is an electromyograph signal.
It can be seen that the maximum amplitude reached
values close to 0.3 mV, while the average signal
value during the attack was about 0.1 mV.
Significant increase in muscle tension with lack of
visible tremors characterizes the tonic phase of the
epilepsy.
5.3.2 Clonic phase
The second part of the tonic-clonic seizure is the
clonic phase. During the test execution, there were
performed some rapid limb movements with
tensioning and relaxing of muscles. The significant
increase in both accelerometer and electromyograph
signals can be observed. For the accelerometer the
average value was over 3.5 m/s2 and the maximum
amplitude was near 20 m/s2. This means very rapid
arm movements. These movements was fast, as the
number of passes by zero is close to Test case 2,
where tremors received smaller accelerometer
values. In the case of the signal from the
electromyograph there is also visible considerable
muscle tension. In the case of second and third time
window, the signal did not fall to zero. In addition,
the mean absolute value was 0.1 mV, while the
maximum amplitude was 0.3 mV, indicating that
muscle strain was significant. The clonic phase of
the epileptic seizure is characterized by strong limbs
tremors and significant muscle strain.
Table 7. Statistics of the Test case 4 for ACC
Time window 1 2 3 4
No. of zero
crossings
86 64 56 51
Mean waveform
length [s]
0.05 0.08 0.09 0.10
Mean absolute
value [m/s2]
0.060 2.45 3.86 1.22
Maximum
amplitude [m/s2]
5.70 15.36 19.48 7.45
Median
frequency [Hz]
21.67
14.01
12.51
11.83
Mean frequency
[Hz]
19.93
13.84
12.24
11.60
Mean power 1.53 9.52 16.90 5.89
Peak frequency
[Hz]
19.80
12.87
11.26
10.25
Table 8. Statistics of the Test case 4 for EMG
Time window 1 2 3 4
No. of zero
crossings
44
0
0
1
0
Mean waveform
length [s]
0.38 5 5 0.25
Mean absolute
value [mV]
0.05 0.11 0.10 0.07
Maximum
amplitude [mV]
0.27 0.30 0.22 0.24
Median
frequency [Hz]
13.18 0 0 0
Mean frequency
[Hz]
17.36
0
0
7.68
Mean power
0.04
0
0
0.21
Peak frequency
[Hz]
10.01
0.20
0.20
2.02
-0.2
-0.15
-0.1
-0.05
0
0.05
0.1
0.15
0.2
15:248 15:441 15:635 15:832
-0.15
-0.1
-0.05
0
0.05
0.1
0.15
Tremor [m/s2]
Time [s]
Voltage [mV]
ACC EMG
WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE
E. Korobko, S.Vilanskaya, A. Baeshko, N. Shestak
E-ISSN: 2224-2902
142
Volume 14, 2017
Fig. 10. Graph of extracted 150 samples started at 1000th
sample of the Test case 4. Blue line is an accelerometer
signal, orange line is an electromyograph signal.
6 Summary
Presented in the paper method and tool offers
proprietary algorithms to support acquisition,
analysis and medical diagnostics of selected
neurological symptoms. The mechanisms cannot
replace the professional help and detailed medical
diagnostics. During system tests implemented
algorithms proved their usefulness especially in the
range of tremors intensity evaluation. Such
information can be used for PD’s drug usage
forecast, as the pharmaceuticals have specific action
time.
many but can assist clinical tests, but can be helpful
when Application of sensors and complex
algorithms requires specific configuration and
environment preparation.
Results received during tests were consisted with
theoretical expectations. Signals that were picked up
by the application from the armband sensor
reflected the behaviour of the tested person. With
increased muscle tension, electromyographic signal
significantly increased its value and vary depending
upon the strength of the tension. A similar situation
was observed in the accelerometer signal. Values
were changing with the movement of the limb and
band sensor obtained values corresponded to
smartphone accelerometer signal. This allows to
state that data received from the band were correct.
In addition, results of tests were consistent with
the assumptions. On the basis of signals’
characteristics, algorithm properly diagnosed
diseases. This allows to evaluate that produced
diagnostic tool has been implemented correctly. It is
a good base for further development or research
related to the diagnosis of neurological disorders
with mobile devices. RFID technology [14][15]
proved to be useful channel for clinical test initial
configuration and on-site clinical test collection
mechanisms. It has been used to authenticate and
preconfigure transmission channel while
maintaining patient’s clinical records anonymous.
Further extensions of RFID usage for automatic
drug dose configuration and drug repository
maintenance.
What is more, the way the application was built
allows to expand it with another measurement
devices or to implement more complex mechanisms
for inferring and analysing user behaviour.
The system has been officially deployed after
successful beta tests on population of 10 patients
which have been using the application for at least 4
months. At the time of paper submission a formal
acceptance of system and sensor clinical tests have
been approved.
References:
[1] M. Nowotarski, Opracowanie algorytmów
decyzyjnych wykorzystujących wybrane
sygnały inercjalne oraz biomedyczne do
identyfikacji zaburzeń neurologicznych oraz
oprogramowania mobilnego na platformę
Android, master’s thesis, M. Chmielewski
(supervisor), MUT (2016),
[2] R. LeMoyne, Wearable and wireless
accelerometer systems for monitoring
Parkinson’s disease patients—A perspective
review, Advances in Parkinson’s Disease, 2,
113-115 (2013)
[3] S.H. Roy, B.T. Cole, L.D. Gilmore, C.J. De
Luca, C.A. Thomas, M.M. Saint-Hilaire, S.H.
Nawab, High-Resolution Tracking of Motor
Disorders in Parkinson’s Disease During
Unconstrained Activity, Movement Disorders,
28, 1080-1087 (2013)
[4] S. Beniczky, T. Polster, T. W. Kjaer, H.
Hjalgrim, Detection of generalized tonic–clonic
seizures by a wireless wrist accelerometer: A
prospective, multicenter study, Epilepsia, 54,
58-61 (2013)
[5] M.M. Velez, R.S. Fisher, V. Bartlett, S. Le,
Tracking generalized tonic-clonic seizures with
a wrist accelerometer linked to an online
database, Seizure, 39, 13-18 (2016)
[6] J.R. Villar, M. Menendez, J. Sedano, E. de la
Cal, V.M. Gonzalez, Analyzing accelerometer
data for epilepsy episode recognition, 10th
International Conference on Soft Computing
Models in Industrial and Environmental
Applications. Advances in Intelligent Systems
and Computing, 368, 39-48, Springer, Cham
(2015)
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
20:232 20:461 20:656 20:854
-18
-13
-8
-3
2
7
12
17
Tremor [m/s2]
Time [s]
Voltage [mV]
ACC EMG
WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE
E. Korobko, S.Vilanskaya, A. Baeshko, N. Shestak
E-ISSN: 2224-2902
143
Volume 14, 2017
[7] SENSE project Wiki,
http://uranus.wat.edu.pl:8808/wiki/index.php/
SENSE (access: 2017.05.05)
[8] PATRON project wiki,
http://uranus.wat.edu.pl:8808/wiki/index.php/
PATRON (access: 2017.05.05)
[9] Myo tech spec: https://www.myo.com/techspecs
[10] B. Stern: Inside Myo, adafruit.com (2016),
https://learn.adafruit.com/myo-armband-
teardown/inside-myo
[11] Myo Bluetooth Low Energy specification file,
https://github.com/thalmiclabs/myo-
bluetooth/blob/master/myohw.h
[12] R. Hoffmann, M. Kiedrowicz, J. Stanik, Risk
management system as the basic paradigm of
the information security management system in
an organization, 20th International Conference
on Circuits, Systems, Communications and
Computers, MATEC Web of Conferences,
Cedex A: E D P Sciences, vol. 76, (2016).
[13] M. Kiedrowicz, T. Nowicki, R. Waszkowski,
Z. Wesolowski, and K. Worwa, Method for
assessing software reliability of the document
management system using the RFID
technology, 20th International Conference on
Circuits, Systems, Communications and
Computers, MATEC Web of Conferences,
Cedex A: E D P Sciences, vol. 76, (2016).
[14] M. Kiedrowicz, Location with the use of the
RFID and GPS technologies - opportunities and
threats, GIS ODYSSEY 2016, pp. 122-128,
(2016).
[15] M. Kiedrowicz, Objects identification in the
informations models used by information
systems, GIS ODYSSEY 2016, pp. 129-136,
(2016).
[16] M. Kiedrowicz, T. Nowicki, R. Waszkowski,
Business process data flow between automated
and human tasks, 3rd International Conference
on Social Science (ICSS 2016) December 9–11
2016, pp. 471-477, (2016).
[17] M. Chmielewski, M. Nowotarski, Sensor-based
mobile system supporting Parkinson disease
clinical tests utilising biomedical and RFID
technologies, 21st International Conference on
Circuits, Systems, Communications and
Computers (CSCC 2017), Agia Pelagia Beach,
Crete Island, Greece, July 14-17, 2017
WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE
E. Korobko, S.Vilanskaya, A. Baeshko, N. Shestak
E-ISSN: 2224-2902
144
Volume 14, 2017
