Intelligent Fall Prevention for Parkinson’s Disease Patients based on
Detecting Posture Instabilily and Freezing of Gait
Jiann-I Pan and Yi-Chi Huang
Department of Medical Informatics, Tzu-Chi University, Hualien, Taiwan
Keywords: Sensor Network, Parkinson’s Disease, Freezing of Gait, Derivative Dynamic Time Warping Algorithm,
Accelerometer.
Abstract: Parkinson’s disease (PD) is a disorder that affects nerve cells in a part of the brain, and results from a
progressive loss of dopaminergic and other sub-cortical neurons. Symptoms of Parkinson’s disease may
include resting tremor, bradykinesia, rigidity, a forward stooped posture, postural instability, and freezing of
gait. As reported by several researchers, the forward stooped posture and freezing of gait are the most
critical reasons to make the Parkinson’s disease patients fall. The main objective of this research is to
develop a fall prevention system for Parkinson’s disease patients. There are two phases in the fall prevention
protocol. The first phase is to detect and recognize the stooped posture and freezing of gait symptoms from
the patient’s movement activities. The next phase is to alarm an audio cue to break the block of freezing. An
accelerometer based sensor network is designed to sense the movement information. The recorded data are
transferred to the smartphone, which served as the core calculator unit, by Bluetooth communication
protocol. The input signals are recognized and classified into the target symptoms. The main advantages of
this proposed approach includes: (1) the safety: to detect the stooped posture and freezing of gait and to
produce audio cue to help the patients to break the block; (2) the portability: not limited at specific
locations; and (3) the expendability: easy to update or upgrade by using app install online.
1 INTRODUCTION
Parkinson’s disease (PD) is a chronic, progressive
neurodegenerative disease, usually occurs in the
elderly population. The average age of onset is
approximately 60 years old (Bloem, et al, 2004).
Clinical symptoms of Parkinson's disease include
tremor, rigidity, akinesia, bradykinesia, postural
instability, gait blocking, etc. (Voss, et al., 2012).
Parkinson's patients exhibit mainly autonomous
action problems, such as akinesia, be referring to the
difficult initial action, and gait block, means action
to freeze or stop suddenly (which also called
freezing of gait, FOG). According to statistics, with
the progress of disease, almost all patients with
Parkinson's disease are being impact by FOG (Giladi
and Balash, 2005). As studied by (Bloem, et al,
2004), blocking gait and postural instability are the
main reasons for caused fall by Parkinson's disease
patients. The proportion of patients with Parkinson's
disease had a fall of about 38% to 68% (Balash, et al,
2005) (Hiorth, et al., 2013). Therefore, the falls are
such an important risk of Parkinson's disease patient
groups. So how to avoid patient falls is a very
important research issue. Currently in the field of
prevention of falls mostly based health education,
changing the factors will cause the fall in life-
environment of ways to promote (Sherrington, et al.,
2008) (Doughty, 2000) (Duncan, et al., 2012), have
less of information technology (IT) aspects
participated. Most of IT-based approaches are focus
on fall detection while fall event happened. This
result is reasonable because of there are too complex
situations to made the fall unpredictable. It is
difficult to recognize all prognostic fall situations. In
order to make the fall prevention feasible, we focus
on two most important situations to be monitored,
that are gait of freezing and from sit to stand and
walk (short as sit-stand-walk). On the other hand,
according to the clinical studies of (Mak and Hui-
Chan, 2004) (McIntoch, et al., 1997) (Powell et al.,
2010), when Parkinson's disease patients in gait of
freezing, can give visual or auditory stimuli so as to
help the patient break the freezing status.
As mentioned above, the objective of this
research is to develop a fall prevention system for
608
Pan J. and Huang Y..
Intelligent Fall Prevention for Parkinson’s Disease Patients based on Detecting Posture Instabilily and Freezing of Gait.
DOI: 10.5220/0005560506080613
In Proceedings of the 12th International Conference on Informatics in Control, Automation and Robotics (ICINCO-2015), pages 608-613
ISBN: 978-989-758-122-9
Copyright
c
2015 SCITEPRESS (Science and Technology Publications, Lda.)
Parkinson's disease patients. There are two phases in
the fall prevention. The first phase is to detect
patients with gait freezing and posture instability
caused by sitting to standing posture forward
bending. The second phase is using voice
instructions (audio cue) to get patient attention for
breaking the freezing gait. In this system, an
accelerometer-based module is used to detect the
patterns of sit-stand-walk and gait of freezing, and
communicated with the smart phone by blue-tooth.
The prognostic fall patterns are recognized by
Derivative Dynamic Time Warping (DDEW)(Keogh
and Pazzani, 2001) algorithm, and trigger an audio
cue (sound as tick-tack) to break the freezing.
2 RELATED WORK
Bloem et al. (Bloem, et al., 2004) resulting the
postural instability and gait impairment are the most
common posture problems in a fall survey. Postural
instability generally refers to the process of unstable
posture. Especially, stooped posture and from sit to
stand and begin to walk are the most prone to fall;
on the other hand, in respect of gait impairment,
freeing the gait is the most directly relevant with
falls. Parkinson Disease Foundation made a similar
recommends to patients with Parkinson's disease
when changing from a sitting position to stand up
and move forward should pay special attention to
avoid out of balance and fall. Freezing of gait is
particularly likely to occur in the beginning from sit
to stand and to walk.
Morris et al. (Morris et al., 1996) found that
using a metronome or a fixed beat music as the
auditory stimulation on gait parameters in patients
with PD has overall improvement. For examples, in
gait frequency and stride length increased by 10% to
20%, further walking speed is increased to 35% to
40%. Mak et al. (Mak and Hui-Chan, 2004) prove
that such feed-forward signals improving the PD
patients impaired nervous system stimulation with
enhanced effect.
In typically, there are four sensor-based methods
to detect the early of freezing and the onset of
freezing. The first method is to use a video camera
to do image analysis (Chen et al., 2011) (Hubble et
al., 1993) (Lozano, et al., 1995). This approach
analysis the body posture of the patient, leg bending
amplitude, and the pace length. The second method
is to measure the lower extremity muscle strength
(Bovi et al., 2011) (Nieuwboer, et al., 2004). This
approach uses EMG Chart to analysis lower limb
muscles condition of the patient. The third method is
to measure foot pressure, using special trail of
sensing foot trampling pressure (Leddy, et al., 2011).
This approach allows patients to walk on the trail,
and execute gait analysis based on the pressure
distribution. Hardware devices and equipment
required for the above-mentioned three methods
have a certain volume and cannot be moved, and
thus limiting the range of motion of subjects and
testing time. The fourth method is to use wireless
sensor (three-axis accelerometer, gyroscope) worn
on patients in different locations (Sant’Anna, et al.,
2011) (Bächlin, et al., 2010) (Godfrey et al., 2011)
(Lee et al., 2010). The collected data is transmitted
to the computing unit processing. This method is
relatively unrestricted in the practice, because the
hardware device is lightweight and easy to wear off.
With MEMS technology matures, wireless
sensors including accelerometers, gyroscopes,
instruments, has been widely used in research of
body movement detection, gait analysis. Sensor
measurement accuracy often depends on the location
and number of worn, configure the location more,
larger sensor number is used, its data will be more
and more accurate. However, the effectiveness and
cost considerations in the central processor, the
number of sensors must be maintained at a certain
amount and the most critical position. Participants
usually placed the triaxial-accelerometer sensors
around the right and left thigh and wrist, and/or
placed triaxial-accelerometers and gyroscopes at
waist and chest.
3 MATERIAL
The purpose of this research is to detect the
imbalanced posture and gait freezing occurs as early
as possible for Parkinson's patients, and then use
audio cues to break the freezing to achieve the goal
of prevention of falls. Although fall is a very
complex behaviour, it is difficult to define all
possible situations. Not all causes of falling can
clearly identify their model (patterns), but if the
results of this research can reduce the risk of falls in
patients with Parkinson's, patients, it can have an
important contribution to patients, their families and
society.
Figure 1shows the proposed system architecture.
The patient’s activity acceleration data are collected
by the sensor which communicated with smartphone
by Bluetooth. The smartphone is used to collect part
of activity data but also play the main processing
work to recognize the freezing pattern and trigger an
alarm.
IntelligentFallPreventionforParkinson'sDiseasePatientsbasedonDetectingPostureInstabililyandFreezingofGait
609
Figure 1: The proposed system architecture.
3.1 Gait Analysis
In this experiment, a gait unit is defined as a gait
cycle which contains space and time parameters. In a
normal gait cycle is started from a referenced
supporting heel touching the ground, and terminated
by the same referenced heel touches the ground
again. In some abnormal gait, the heel may not be
the first part of the foot touches the ground.
Therefore, this type of gait cycle should be
considered to refer to some part of the foot touches
the ground start, until the same part again touching
the ground for termination.
Gait cycle can be divided into two phases, the
stance phase and the swing phase. In normal gait, the
stance phase accounts for the 60% of entire gait
cycle, which is defined as a referenced limb foot
ground contact period. On the other hands, swing
phase accounts for the 40% in the gait cycle, which
is defined as the referenced limb foot without
touching the ground during this time. For example,
when the right (left) leg is in the stance phase, the
left (right) leg will be in the swing phase. Therefore,
a gait cycle includes right and left of the stage. A
single swing phase of gait cycle includes the right
side and the left side.
The double support time refers the time of the
body weight is transferred from one foot to another
foot, i.e. both left and right foot during ground
contact simultaneously in one gait cycle. These
variables can be used to measure a gait cycle: the
time of standing (right and left), the time of leaving
the ground (right and left), the time of double
support, and the whole gait time.
3.2 Accelerometer-based Sensor
An accelerometer-based sensor was designed to
capture body acceleration data. The sensor
comprises the following components: (1) a
LIS3LV02DQ triaxial-acceleration sensing unit, (2)
a MSP430F169 microcontroller processing unit, (3)
a BTM-112 Bluetooth module wireless transmission
unit, and (4) a lightweight rechargeable lithium
battery as the power supply unit. The sensor
dimensions are 40 mm * 28 mm * 18 mm (see
Figure 2). The sampling frequency was set as 32 Hz.
Figure 2: The accelerometer-based sensor.
3.3 Placement of Sensors and
Smartphone
The accelerometer-based sensors are placed on the
left and right ankle (as the Figure 3 shown), and
smartphone is placed on the waist (as the Figure 4
shown).
Figure 3: The accelerometer-based sensors are placed on
the left and right ankle.
Figure 5 shows a sample acceleration data of a
patient from sit to stand and walk which captured by
the smartphone (HTC Desire A8181, Android-based
smartphone).
Figure 6 shows part of the 3-axis acceleration
data of freezing of gait from ((Bächlin, et al., 2010).
ICINCO2015-12thInternationalConferenceonInformaticsinControl,AutomationandRobotics
610
Figure 4: The smartphone is placed on the waist.
Figure 5: An acceleration data of sit-stand-walk activity
captured from the smartphone.
Figure 6: Part of tri-axis acceleration data of freezing of
gait.
3.4 System Flow
The smartphone plays the fall prevention decision
centre, which receives the tri-axial acceleration data
from ankle sensors and smartphone built-in
accelerometer. Figure 7 presents the system flow of
fall prevention process.
Figure 7: The system flow.
There are four phases in the proposed approach.
First, the captured three-axis acceleration sensor data
from the ankle transmitted via wireless Bluetooth to
smartphone, and integrated with the acceleration
data captured by the waist smartphone built-in
accelerometer. In the next phase, according to the
triaxial accelerometer signals probably mixed some
noise. Therefore, a nonlinear median signal filter is
used to filter noise. As a sliding window by
capturing the odd samples, select the sort to replace
the original value in the middle of the sample values.
The filtering effect is related with the window size
used. In preliminary analysis of this study, window
size is set as 7.
In the third phase, the derivative DTW algorithm
(DDTW) is adapted as the primary algorithm for
determining the fall auspice. The DDTW is extended
from classical DTW. DTW is a conventional method
of alignment similarity of two time series,
commonly used in speech recognition and action
recognition. Since the length ratio of the two time
series may be inconsistent, the use of Euclidean
IntelligentFallPreventionforParkinson'sDiseasePatientsbasedonDetectingPostureInstabililyandFreezingofGait
611
distance cannot correctly calculate the distance
between the two sequences (and similarity). As such,
DTW algorithm calculates the sequence similarity
by extending or shortening the sequence length. For
this purpose, the timeline sequence will exhibit the
phenomenon of non-linear and warp, so that
corresponding points of two sequences hold time
consistency. That is, the best warp-path can be
defined. However, the traditional DTW algorithm
sometimes has misalignment results. When multiple
points on a time series to the same single point,
called singularities, or the two time series have large
difference on Y axis are likely to make mistakes.
Therefore, DDTW proposed amendment for
modified X-axis before the Distance matrix
calculated.
The main reason of using DDTW is to consider
each gait in patients with symptoms of the freeze of
its time series may not be the same, reliable
assessment can be obtained through the DDTW
algorithm. The captured three-axis acceleration
serial signals are comparing with the postural
instability and freezing of gait reference model
established from the DDTW analysis. The three-axis
acceleration data of postural instability and freezing
of gait reference model is constructed from three
sources: (1) Parkinson's patient gait freezing
acceleration data set, built by the University of
California (Bach and Lichman, 2011)(Bächlin, et
al.,2010); (2) simulation data of posture and gait in
patients with Parkinson's symptoms of the freeze by
healthy subjects took the proposed accelerometer
and smartphone; and (3) real acceleration data of
posture and gait from 3-5 real Parkinson’s patient.
The final phase is to trigger an audio cue using
the build-in speaker in smartphone. The cue will
terminate if the subject can move continued (i.e.,
return to normal gaits) after audio cue 10 seconds, or
otherwise, the smartphone will contact the family or
others.
3.5 Experiments
The experiment is currently in progressing, which
approved by Research Ethic Committee of Hualien
Tzu-Chi Hospital, Taiwan (IRB103-67-B).
Sample data category is expected to include the
following categories: from sit to stand and walk
stooped posture, FOG episodes precursor and FOG
episodes (includes gait cycle disorder, such as split
step, walking speed decreases, stop walking, etc.),
and normal stop.
The evaluation of the proposed system will
include the following three parts:
(1). evaluate the recognition rate: it is mainly to
test the accuracy of identification of FOG and sit-
stand-walk patterns, including the correct and
incorrect classification.
(2). evaluate the audio cue issued latency:
although there is no formal clinical empirical time
analysis of fall events in the end occurs after the gait
freeze happen. But the sooner the issue is better
audio cue.
(3). verify that the overall accuracy of the system
functionality, including follow audio cue
termination, notify relatives and other functional
requirements.
More detailed discussion will be given after the
experiments completed.
4 CONCLUSIONS
The main objective of this research is to develop a
fall prevention system for Parkinson’s disease
patients. There are two phases in the fall prevention
protocol. The first phase is to detect and recognize
the stooped posture and freezing of gait symptoms
from the patient’s movement activities. The next
phase is to alarm an audio cue to break the block of
freezing. An accelerometer based sensor network is
designed to sense the movement information. The
recorded data are transferred to the smartphone,
which served as the core calculator unit, by
Bluetooth communication protocol. The input
signals are recognized and classified into the target
symptoms. The main advantages of this proposed
approach includes: (1) the safety: to detect the
stooped posture and freezing of gait and to produce
audio cue to help the patients to break the block; (2)
the portability: not limited at specific locations; and
(3) the expendability: easy to update or upgrade by
using app install online.
ACKNOWLEDGEMENTS
This project was funded by the Ministry of Science
and Technology, Taiwan (grant number MOST 103-
2221-E-320 -002).
REFERENCES
Amboni, M., Barone, P., Iuppariello, L., et al, 2012. Gait
Patterns in Parkinsonian Patients With or Without
ICINCO2015-12thInternationalConferenceonInformaticsinControl,AutomationandRobotics
612
Mild Cognitive Impairment, Movement Disorders,
27(12), pp.1536-1543.
Sant’Anna, A., Salarian, A., and Wickstrom, N., 2011. A
New Measure of Movement Symmetry in Early
Parkinson’s Disease Patients Using Symbolic
Processing of Inertial Sensor Data, IEEE Transactions
on Biomedical Engineering, 58(7), JULY, pp.2127-
2135.
Bache, K. and Lichman, M., 2013. UCI Machine Learning
Repository [http://archive.ics.uci.edu/ml]. Irvine, CA:
University of California, School of Information and
Computer Science.
Bächlin, M., Plotnik, M., Roggen, D., Maidan, I.,
Hausdorff, J. M., Giladi, N. and Tröster G., 2010.
Wearable Assistant for Parkinson's Disease Patients
with the Freezing of Gait Symptom, IEEE
Transactions on Information Technology in
Biomedicine, 14(2), pp 436-446.
Balash, Y. et al. 2005. Falls in outpatients with Parkinson's
disease. Journal of Neurology Vol.252, pp.1310-1315.
Bloem, B. R., Hausdorff, J. M., Visser, J. E., and Giladi,
N. 2004. Falls and freezing of gait in Parkinson’s
disease: a review of two interconnected, episodic
phenomena, Movement Disorders 19(8), August,
pp.871–884.
Bovi, G., Rabuffetti, M., Mazzoleni, P., and Ferrarin,M.
2011. A multiple-task gait analysis approach:
Kinematic, kinetic and EMG reference data for healthy
young and adult subjects, Gait & Posture Vol.33, pp.
6–13.
Chen, S. W., Lin, S. H. Liao, L. D. et al., 2011.
Quantification and recognition of parkinsonian gait
from monocular video imaging using kernel based
principal component analysis, BioMedical
Engineering OnLine, 10:99.
Dibble L. E., Christensen J., Ballard D. J. and Foreman K.
B., Diagnosis of Fall Risk in Parkinson Disease: An
Analysis of Individual and Collective Clinical Balance
Test Interpretation, PHYS THER. Vol.88, 2008,
pp.323-332.
Doughty K., 2000. Fall prevention and management
strategies based on intelligent detection, monitoring
and assessment. Presented at New Technologies in
Medicine for the Elderly, Charing Cross Hospital,
30th, Nov.
Duncan, R. P. Leddy A. L., Cavanaugh J. T., Dibble L. E.,
Ellis T. D., Ford M. P., Foreman K. B., and Earhart
G.M., 2012. Accuracy of Fall Prediction in Parkinson
Disease: Six-Month and 12-Month Prospective
Analyses. Parkinson’s Disease. Article ID 237673,
doi:10.1155/2012/237673.
Giladi, N. and Balash, Y., 2005. "Treatment of
Parkinsonian gait disturbances." Evaluation and
management of gait disorders. Marcel Dekker Inc., pp.
273-83.
Godfrey A., Bourke, A. K., Olaighin, G. M., van de Vend,
P., Nelsond, J., 2011. Activity classification using a
single chest mounted tri-axial accelerometer, Medical
Engineering & Physics Vol.33, pp. 1127– 1135.
Hiorth, Y. H., Lode, K. and Larsen, J. P., 2013.
Frequencies of falls and associated features at different
stages of Parkinson’s disease. European Journal of
Neurology Vol.20, pp.160–166.
Keogh, Eamonn J., and Michael J. Pazzani. "Derivative
Dynamic Time Warping." SDM. Vol. 1. 2001.
Leddy, A. L., Crowner, B. E. and Earhart, G. M., 2011.
Functional Gait Assessment and Balance Evaluation
System Test: Reliability, Validity, Sensitivity, and
Specificity for Identifying Individuals With Parkinson
Disease Who Fall, Physical Therapy 91(1), January,
pp.102-113.
Lee, J. A., Cho, S. H., Lee, Y. J., Yang, H. K., Lee, J. W.,
2010. Portable, Portable Activity Monitoring System
for Temporal Parameters of Gait Cycles, J Med Syst,
Vol.34:959–966.
Mak, M. K. Y., Hui-Chan, C. W. Y., 2004. Audiovisual
cues can enhance sit-to-stand in patients with
Parkinson's disease, Movement Disorders, 19(9):
1012-1029.
McIntosh, G. C., Brown, S. H., Rice, R. R., Thaut, M. H.,
1997. Rhythmic auditory-motor facilitation of gait
patterns in patients with Parkinson's disease, Journal
of Neurology, Neurosurgery, and Psychiatry, Vol.
62:22-26.
Morris, M. E., Iansek, R., Matyas, T. A., Summers, J. J.,
1996. Stride length regulation in Parkinson’s disease
normalization strategies and underlying mechanisms.
Brain Vol.119, pp.551–568.
Nieuwboer, A., Dom, R., De Weerdt, W., et al., 2004.
Electromyographic profiles of gait prior to onset of
freezing episodes in patients with Parkinson’s disease,
Brain, vol.127, 1650b–1660b.
Powell, W., Stevens, B., Hand, S. and Simmonds, M.,
2010. Sounding Better: Fast Audio Cues Increase
Walk Speed in Treadmill-Mediated Virtual
Rehabilitation Environments, School of Creative
Technologies, University of Portsmouth, UK, School
of Physical & Occupational Therapy, McGill
University, CANADA.
Sherrington, C., Whitney, J. C., Lord, S. R., Herbert, R.
D., Cumming, R. G., and Close, J. C. T., 2008.
Effective Exercise for the Prevention of Falls: A
Systematic Review and Meta-Analysis. Journal of the
American Geriatrics Society 56(12): 2234-2243.
Voss, T.S., Elm, J. J., Wielinski, C. L., Aminoff, M. J.,
Bandyopadhyay, D., Chou, K. L., Sudarsky, L. R.,
Tilley, B. C., 2012. Fall frequency and risk assessment
in early Parkinson’s disease, Parkinsonism and
Related Disorders 18, pp.837-841.
IntelligentFallPreventionforParkinson'sDiseasePatientsbasedonDetectingPostureInstabililyandFreezingofGait
613
