gait feature vectors for post-stroke prediction using

Gait Feature Vectors for Post-stroke Prediction using Wearable Sensor

Seunghee Hong1, Damee Kim2, Hongkyu Park3, Young Seo4,

Iqram Hussain5 Se Jin Park6†

Abstract

Stroke is a health problem experienced by many elderly people around the world. Stroke has a devastating effect

on quality of life, causing death or disability. Hemiplegia is clearly an early sign of a stroke and can be detected

through patterns of body balance and gait. The goal of this study was to determine various feature vectors of foot

pressure and gait parameters of patients with stroke through the use of a wearable sensor and to compare the gait

parameters with those of healthy elderly people. To monitor the participants at all times, we used a simple measuring

device rather than a medical device. We measured gait data of 220 healthy people older than 65 years of age and

of 63 elderly patients who had experienced stroke less than 6 months earlier. The center of pressure and the

acceleration during standing and gait-related tasks were recorded by a wearable insole sensor worn by the participants.

Both the average acceleration and the maximum acceleration were significantly higher in the healthy participants

(p < .01) than in the patients with stroke. Thus gait parameters are helpful for determining whether they are patients

with stroke or normal elderly people.

Key words: Gait Analysis, Health Monitoring System, Post-Stroke, Wearable Sensors

1. Introduction1)

The World Health Organization (WHO) defines

stroke as A stroke is caused by the interruption of the

blood supply to the brain, usually because a blood ves-

sel bursts or is blocked by a clot. This cuts off the sup-

ply of oxygen and nutrients, causing damage to the

brain tissue. Strokes are classified as ischemic stroke

which caused by inadequate supply of oxygen and nu-

trients in the blood after cerebrovascular blockage, and

Hemorrhagic stroke which caused by hematoma caused

by cerebral vascular injury. Risk factors of stroke onset

are known to be related to hypertension, diabetes, obe-

sity, dyslipidemia, etc. The Stroke Association recom-

1 Seunghee Hong: Senior Researcher, Center for Medical Convergence Metrology, KRISS2 Damee Kim: Researcher, Center for Medical Convergence Metrology, KRISS;

Researcher, Department of KSB Convergence Research, ETRI3 Hongkyu Park: Researcher, Department of KSB Convergence Research, ETRI4 Young Seo: Student Researcher, Center for Medical Convergence Metrology, KRISS5 Iqram Hussain: Student Researcher, Center for Medical Convergence Metrology, KRISS;

Student, Medical Physics, UST6 Se Jin Park: Head Researcher, Center for Medical Convergence Metrology, KRISS;

Head Researcher, Department of KSB Convergence Research, ETRI

Professor, Medical Physics, UST

†(Corresponding Author) Se Jin Park: Center for Medical Convergence Metrology, KRISS / E-mail：[email protected] /

TEL：042-868-5450

감성과학제22권 3호, 2019<연구논문>

pISSN 1226-8593eISSN 2383-613X

Sci. Emot. Sensib.,Vol.22, No.3, pp.55-64, 2019

https://doi.org/10.14695/KJSOS.2018.22.3.55

56 Seunghee Hong․Damee Kim․Hongkyu Park․Young Seo․Iqram Hussain․Se Jin Park

mends that it is important to identify and track patho-

physiologic risk factors with appropriate diagnostic

tests for recurrence and secondary prevention of stroke.

However, according to the Global Burden of Disease

(GBD) survey, the mortality rate of stroke among the

elderly over 70 years old was 7.76% in USA, 6.99%

in Germany, 9.47% in Japan and 13.02% in Korea

(GBD, 2017). It is difficult to directly compare the dif-

ference in race, culture, economic level and medical

level according to the region, but in Korea, the mortal-

ity rate due to stroke is significantly higher than in oth-

er countries. Symptoms and the severity after stroke

are determined by the position of the brain lesion and

the size of the lesion, but it is most important to treat

the specialist immediately after the onset. Furthermore,

due to this, the timing of appropriate treatment is

missed, increasing the patient's mortality rate and the

rate of disability. The typical clinical features of stroke

include hemiplegia, disturbance of sense, spasticity, in-

coordination, cognitive impairment, apraxia, hemi-

neglect, and dysphagia. In the early stages of stroke,

muscle weakness and decreased muscle tension lead to

paralysis. Among them, hemiplegia due to neuro-

logical disorders is very common in more than 80% of

stroke patients in general (Duncan et al., 2015).The

weakening of neurological muscular strength makes it

difficult for stroke patients to maintain the center of

the body in a static standing situation or to make nor-

mal walking (Duncan et al., 2015; Wevers et al.,

2009). The movement of the pressure center (center of

pressure, CoP) of the body in the static standing pos-

ture is being used as an evaluation tool to diagnose the

balance disorder (Bohannon, 1997). Gollhofer (1989)

and Horstmann (1990) reported that the trajectory of

the center of gravity (COG) gives information on the

control of the movements of the body necessary to

maintain a standing posture. While the trajectory of

difference of the center of pressure from the center of

gravity (COP-COG) presents the neuromuscular stiff-

ness of the biomechanical system controlling the center

of gravity. Moreover, Foot Pressure is and index to

evaluate the parameter of balance and walking. It may

cause musculoskeletal damage as well as physiological

disorders when foot pressure exceeds the normal range

(Yang et al., 2015).

Articulation in walking after a stroke involves a co-

operative pattern of lower limbs rather than selective

movement (Chen et al., 2005). Then, it adapts to a saf-

er and more stable walking pattern in order to magnify

the decrease of sensory information from the foot

(Yang et al., 2005). Insufficient single-leg support or

uncontrolled forward movement is considered a funda-

mental dysfunction leading to asymmetry. Asymmetry

consists of a decrease in stance period time and ex-

tension of swing leg period (Peter & Arthur, 2005).

Both leg support periods and stance periods are longer

than normal on both legs (Bohannon, 1986; Bohannon,

1987; Bohannon, 1991; Bohannon, 1997; Brandstater et

al., 1983; Nadeau et al., 1999; Nakamura et al., 1998).

Such asymmetry leads to an increase in energy costs,

falls, joint abnormalities, burden on joints, and pain

(Morris et al., 1986; Olney, 1969). Methods for meas-

uring such asymmetry are described in various

documents. The expression for obtaining temporal

asymmetry is calculated by dividing the value obtained

at the paralyzed side swing period / paralyzed side

stance period by the non-paralyzed side swing period /

non paralyzed side stance period time (Olney, 1969;

Patterson et al., 2010). The asymmetry of the single-leg

support time is obtained by 1-single-leg support time

(paralyzed side) / single leg support time (non-para-

lyzed side). The stride ratio is calculated by dividing

the stride (meters) on the paralyzed side by the non-

paralyzed side stride (meters) (Balasubramanian et al.,

2007). The larger these values, the more notable the

asymmetry is (Hsu, 2003). In stroke patients, the walk-

ing speed generally decreases. In a healthy subject, the

comfortable walking speed is about 1.3 m/s (Bohannon

Gait Feature Vectors for Post-stroke Prediction using Wearable Sensor 57

et al., 1997), and in a hemiplegic patient, the average

walking speed is set to 0.23 - 0.73 m/s (Olney &

Richards, 1996). Perry is classified into three according

to walking speed, and when it is less than 0.4 m / s

it is restricted to indoor movement, some restriction

occurs outdoors at 0.4 to 0.8 m / s, outdoor movement

is not restricted at 0.8 m / s or more It revealed that

(Perry, 1995).

According to the results of previous studies, body

balance and walking characteristics are obvious ob-

stacles caused by stroke. As mentioned above, quanti-

tative evaluation methods by the nervous system dam-

age after stroke onset can be measured, there aren’t

any systems monitoring and detecting the stroke fea-

tures when they are standing up and walking. In other

words, the delay in detection of a sympathetic post-

poned the treatment time of the stroke patients, thus

causing a fatal disorder or affecting the maintenance

of life.

Nevertheless, elder people have been still exposed

to the high risk of stroke should be monitored during

their daily activities and also the sensors and the deci-

sion making algorithms to detect their stroke symp-

toms are needed. At that reason, in this study, the dif-

ference in walking and standing characteristics be-

tween normal elderly and stroke patients is defined as

a feature vectors. And also, these results will be used

as a feature of the algorithm development that will be

applied to the stroke monitoring system during the

walk.

2. Methods

All experimental procedures were carried out after

approval of the institutional review board (KRISS-

IRB-2017-07).

2.1. Stroke Monitoring System

Our elderly health (stoke) monitoring system was de-

signed and suggested as shown in the Fig. 1 (Park et

al., 2017). Hyper-connected self-machine learning en-

gine controls the system. The elements of the system

are a knowledge base, big data, network security, re-

al-time data monitoring using wearables, and self-learn-

ing engine. The knowledge base would have risk fac-

tors, medical health records, psycho-logical factors, gait

Fig. 1. Components of the proposed elderly health (stoke) monitoring system


Fig. 2. System Framework for Stroke monitoring system during walking with wearable sensors

and motion patterns, and bio-signals. The old peoples’

activities, physiological, and bio signals are monitored

in real-time through wearable sensors. The self-ma-

chine learning engine would include multi-model learn-

ing (SVM, Bayes, RF, CNN, LSTM, deep learning)

and model generator. If the proposed system predicts

stroke symptom above 90%, it will generate an alarm

to family, the victim, people around the victim, and

healthcare professionals. Then the victim will get the

timely medical assistance. The Stroke Monitoring

System using gait parameters such as, insole foot pres-

sure and accelerometer, is proposed. Gait patterns such

as gait speed, foot pressure etc. of normal person and

stroke patient are significantly different from each

other. Wearable IoT (Internet of Things) devices and

statistical technique are able to detect stroke onset of

elderly adults. Entire system will feed subject’s physio-

logical data to cloud engine to compare real-time data

and reference data in order to detect stroke during gait

activities.

2.2. Subjects

220 normal elderly people over 65 years old and 63

elderly patients who were less than 6 months after the

onset of stroke participated in this study as in the Table

1. Study was performed in A National University

Hospital, Daejeon, South Korea. Healthy participants

were recruited to the elderly enrolled at the local silver

center, and patients were recruited for the elderly who

were in hospital or after rehabilitation. The stroke pa-

tients were limited to those who could walk more than

30m independently without a walking frame. All partic-

ipants performed basic physical examinations (height,

weight, electrocardiogram, blood pressure, blood test) be-

fore the experiment, and participants were paid partic-

ipation fee including transportation expenses. In addition,

the participants were instructed to give up the experiment

at any time according to the health condition of the ex-

periment day Proposed Gait monitoring system is con-

sists of insole foot Pressure sensor and accelerometer

which will be attached to foot as shoe insole for gather-

ing gait speed, foot pressure and other gait signals.


Normal elderly (n=220) Elderly Stroke Patients (n=63)

Gender (Years)M: 98 (44.5%) F: 122 (55.5%)

M: 42 (66.7%) F: 21 (33.3%)

Age (Mean, std.) 68.45 ± 4.22 73.26 ± 6.18

Height/Weight (cm) 157.13/59.76 157.47/60.19

Systolic /Diastolic Blood pressure (mmHg) 134.22/79.33 134.78/79.55

Table 1. Demographic characteristics of recruited elderly normal and stroke patients

2.3. Data acquisition system

The insole sensor was the Dynamic pressure mapping

system, Dynafoot 2 (TechnoConcept, Pitaugier, MANE,

France). There are 58 pressure (resistance) sensors and

accelerometers on the bottom of the insoles. The thick-

ness of the insole is 2mm and the size of the resistance

sensor is 9 * 9mm and the measurement range is 2000g.

The accelerometer was twin-axial and the measuring

range was +/- 6g. The sampling rate was 100Hz. And

the data transfer module is 49 * 38 * 19mm which is

mounted on the shoe.

In the recording mode, there is no limit to the meas-

uring distance and it is possible to measure up to 3.5

hours.

2.4. Data analysis

Experimental data was measured using Dynafoot 2

and extracted using Dynafoot2_V2.2.14.0 software. The

measured features were the pressure per cell, the center

of pressure [CoP; CoP_Left_x axis (medial, lateral),

CoP_Left_y axis (anterior, posterior), CoP_Right_ x ax-

is (medial, lateral), CoP_Right_ y axis (anterior, poste-

rior)], the acceleration (Acc._left foot, acc._right foot,),

and the ground reaction force (GRF; GFR._left foot,

GFR._ right foot,).

Gait data were analyzed using IBM SPSS Modeler

18.0. After data preprocessing (outlier removal), patient

data were amplified to adjust the data distribution ratio of

stroke patients and normal persons. Then statistical analy-

sis method was paired t-test at a confidence level of 95%.

3. Results

3.1. Static Standing Position

3.1.1. Plantar Center of Pressure (CoP)

CoP displacement was measured in both feet of a

normal elderly person and a stroke patient in a static

standing state, and was analyzed into the medial, later-

al, anterior, and posterior regions (confidence level of

95%) as shown Fig. 3.

3.2. Gait/Movement

3.2.1. Gait task duration

After standing in static condition, participants per-

formed 20m gaiting test along the guide tape line. Fig.

4 shows the gaiting time of the normal group and the

patient group. The normal group reached the end point

with 18.98 ± 4.672s, while the patient group reached

the end point with 43.75 ± 29.830s. There was a sig-

nificant difference in arrival time between two groups

(p-value <.01).

3.2.2. Gait accelerations

The maximum and average accelerations of the feet

were measured from the accelerometer embedded in the

insole, and paired t-test was performed. In the Fig. 5,

the average acceleration of the normal group was meas-

ured as 17.18 ± 10.563m/s² and 19.44 ± 19.365m/s²,

left foot and right foot respectively, and the average ac-

celeration of the patient group was measured as 12.20

± 1.167m/s² and 12.67 ± 1.519m/s². It was clear that


(a) (b)

(c) (d)

Fig. 3. CoP displacement from plantar center of normal group and stroke patients group: (a) medial and lateral CoP displacement of left plantar, (b) anterior and posterior CoP displacement of left plantar, (c) medial and lateral CoP displacement of right plantar, (d) anterior and posterior CoP displacement of right plantar (** p-value <.01)

Fig. 4. Gait duration from the starting point to the end point

(** p-value <.01).

(a) (b)

Fig. 5. Acceleration of the normal group and the stroke patient group during gating: (a) average acceleration, (b) maximum acceleration (** p-value <.01).


the average acceleration of the normal group was sig-

nificantly higher than the average acceleration of the pa-

tient group. The maximum acceleration of the normal

group was 75.24 ± 9.758m/s² and 73.39 ± 11.700m/s²,

left foot and right foot respectively. The maximum ac-

celeration of the normal group was 66.96 ± 11.597m/s²

and 69.28 ± 11.563m/s², respectively. Both the aver-

age acceleration and the maximum acceleration were

significantly higher in the normal group (p-value

<.01).

4. Discussion

The CoP displacement in the left foot of the stroke

patients was significantly larger than that of the normal

elderly. The anterior and posterior CoP displacement

showed a significant increase in stroke patients also.

Especially the anterior movement was very prominent.

It was also found that in the stroke patients, the CoP

of the left foot was shifting in the anterior-medial

direction. On the other hand, in CoP displacement of

the right foot, the shift value in the anterior-posterior

direction was larger in patients with stroke. These re-

sults suggest that stroke patients have difficulty in con-

trolling the balance because they cannot tolerate weight

bearing on the plantar part. Haart et al. (2004) reported

that stroke patients had weight gain in the forefoot due

to an ankle muscle imbalance and that the balance was

shaken to the outside of the foot. It obvious to main-

tain balance in the anterior medial side to control the

unbalance in the stroke patients. Therefore, when the

patient was standing in a static state, it was confirmed

that the stroke patient maintained an anterolateral bal-

ance and that the CoP of the plantar in the ante-

roposterior direction were higher.

The CoP from normal and patient groups were also

plotted and the normal range of pressure center values

was specified. The range of the normal that does not

include outliers was indicated by box in the scatter plot

as shown in the Fig 6. In the left foot, patient data of

the anterior-lateral side and normal group data of the

posterior-lateral side were distributed. On the right

foot, the patient data of the posterior side were dis-

tributed and the data of the normal side were scattered

on the mediolateral part. As a result, CoP values out-

side the normal range were also observed in the normal

group. This suggests that physical degeneration due to

aging has generated an abnormal value even though it

is a normal elderly person.

Fig. 6. CoP distribution of normal elderly people and stroke patient (left and right foot, respectively)


In addition, gait time and acceleration were obtained

from the gait test. The results showed that the gait time

of the normal group was significantly faster and the ac-

celeration value was higher in the normal group. From

these results, it was clear that walking speed was a very

clear feature in predicting stroke. As shown in Fig. 7,

it was confirmed that the acceleration of patients who

have plotted the left and right maximum acceleration

values on a three-dimensional graph was clearly dis-

tinguished from the range of normal persons.

Fig. 7. CoP placement limit box of normal elderly people (left and right foot, respectively)

However, since the shape of the foot is very different

according to the user, and the state of the surface of the

ground is different according to the environment in which

the user lives and the user, it is difficult to detect the ab-

normal data only through the simple statistics. Therefore,

in the future, the conditions for the experimental environ-

ment will be expanded, bio - signal data will be collected,

and reliability will be improved by using algorithms using

machine learning and deep learning.

5. Conclusions

Balance in standing posture and gait in patients with

stroke are important factors in evaluating functional in-

dependence (Turnbull, Charteris, & Wall, 1995). And

those have been analyzed in various methodologies.

Balancing in standing has been evaluated using the

COP, the COP-COG, and the Foot pressure. Hence, the

evaluation tools of the gait are much more diversified.

The common methods for hemiplegic are the velocity,

the cadence, the step length, the single time ratio, and

the stance phase symmetry ratio. Among these, the ve-

locity is easy to measure and corresponds to the clin-

ical condition of the patient (Wall & Turnbull, 1986;

Roth et al., 1997). Also, it was observed that hemi-

plegic patients were closely related to the cadence, the

velocity, the muscle strength, the Barthel index score

and the gait pattern (Dettman, Linder, & Sepic, 1987).

On the other hand, the most prominent gait patterns for

stroke patients presented the low gait velocity, the

short stance phase, and the difference in step length be-

tween the paralyzed side and non-paralyzed side of the

body (Mauritz, 2002). As such, it has been developed

into a multi-faceted evaluation method such as the

COP, COG and foot pressure method when standing

and the velocity, the cadence, the step length, the sin-

gle time ratio, and the stance phase symmetry ratio

method when walking. For real-time health monitoring,

the temporal and spatial variables during daily time

must be measured accurately. In addition, there should

be no limitation in the measurement environment in or-

der to determine the difficulty of maintaining the bal-

ance of the body and the abnormality of walking.

However, foot pressure varies considerably depending

on the road surface, and gyro sensors using magnetic

sensors are affected by such values in environments

with many electronic devices. Therefore, sensors and

measurement variables for daily monitoring should not

be influenced by the environment, and the use of opti-

mal parameters should be a priority.

The purpose of this study was to extract the optimal

variables to detect abnormalities of stationary posture

and walking during daily life, and to find the difference


between stroke patients and the normal elderly. However,

the stroke patients who were in-rolled in this study were

patients within three months of onset and were unable to

collect patient data at the time of onset. However, we

will develop machine learning and deep learning algo-

rithms and embed them in our stroke monitoring system.

After that, we will conduct follow-up studies on older

people to monitor stroke incidence and will also collect

additional data at the time of onset. Once the system has

been developed and validated, it is expected to reduce

the sequelae and mortality caused by delays in treatment

time after stroke.

Acknowledgments

This work was supported by the National Research

Council of Science & Technology (NST) grant by the

Korea government (MSIP) (No. CRC-15-05-ETRI).

REFERENCES

Balasubramanian, L., Ahmed, A., Lo, C. M., Sham, J.

S., & Yip, K. P. (2007). Remanent cell traction force

in renal vascular smooth muscle cells induced by

integrin-mediated mechanotransduction. American

Journal of Physiology-Regulatory, Integrative and

Comparative Physiology, 293(4), R1586-94.

DOI: 10.1152/ajpcell.00234.2012

Bohannon, R. W. (1986). Strength of lower limb related

to gait velocity and cadence in stroke patients.

Physiotherapy Canada, 38(4), 204-206.

DOI: 10.3138/ptc.38.4.204

Bohannon, R. W. (1987). Gait performance of hemiparetic

stroke patients: selected variables. Arch Physical

Medicine and Rehabilitation Clinics of North America,

68(11), 777-781.

Bohannon, R. W. (1991). Strength deficits also predict

gait performance in patients with stroke. Perceptual

and Motor Skills, 73(1), 146.

DOI: 10.2466/pms.1991.73.1.146

Bohannon, R. W. (1997). Comfortable and maximum

walking speed of adults aged 20-79 years: reference

values and determinants. Age and Ageing, 26(1),

15-19. DOI: 10.1093/ageing/26.1.15

Bohannon, R. W. (1997). Gait Performance with Wheeled

and Standard Walkers. Perceptual and Motor Skills,

85(3_suppl), 1185-1186.

Doi:10.2466/pms.1997.85.3f.1185

Brandstater, M. E., de Bruin, H., Gowland, C., & Clark,

B. M. (1983). Hemiplegic gait: analysis of temporal

variables. Archives of Physical Medicine and

Rehabilitation, 64(12), 583-587.

Chen, G., Patten, C., Kothari, D. H., & Zajac, F. E.

(2005). Gait differences between individuals with

post- stroke. Gait Posture, 22(1), 51-56.

DOI: 10.1016/j.gaitpost.2004.06.009

Dettman, M. A., Linder, M. T., & Sepic, S. B. (1987).

Relationship among walking performance, postural

stability and functional assessments of the hemiplegic

patient. American Journal of Physical Medicine and

Rehabilitation, 66, 77-90.

Duncan, P. W., Zorowitz, R., Bates, B., Choi, J. Y.,

Glasberg, J. J., Graham, G. D., Katz, R. C., Lambert,

K., & Reker, D. (2015). Management of adult stroke

rehabilitation care: a clinical practice guideline.

Stroke, 36(9), e100-e143.

DOI: 10.1161/01.STR.0000180861.54180.FF

Haart, D. M., Geurts, A. C., Huidekoper, S. C., Fasotti,

L., & van Limbeek, J. (2004). Recovery of standing

balance in postacute stroke patients: a rehabilitation

cohort study. Archives of Physical Medicine and


DOI: 10.1161/01.STR.0000180861.54180.FF

Hsu, A. L., Tang, P. F., & Jan, M. H., (2003). Analysis

of impairments influencing gait velocity and

asymmetry of hemiplegic patients after mild to

moderate stroke. Arch Phys Med Rehabil, 84(8),

1185-1193. DOI: 10.1016/s0003-9993(03)00030-3


Institute for Health Metrics and Evaluation (IHME),

GBD Compare. Viz Hub

(https://vizhub.healthdata.org/gbd-compare/).

Mauritz K. H. (2002). Gait training in hemiplegia.

European journal of neurology, 9(s1), 23-29.

Doi: 10.1046/j.1468-1331.2002.0090s1023.x|

Morris, R. G., Anderson, E., Lynch, G. S., & Baudry,

M., (1986). Selective impairment of learning and

blockade of long-term potentiation by an N-methyl-

D-aspartate receptor antagonist. AP5. Nature, 27(319),

774-786. DOI: 10.1038/319774a0

Nadeau, S., Arsenault, A. B., Gravel, D., & Bourbonnais,

D. (1999). Analysis of the clinical factors determining

natural and maximal gait speeds in adults with a

stroke. American Journal of Physical Medicine &


DOI: 10.1097/00002060-199903000-00007

Nakamura, R., Watanabe, S., Handa, T., & Morohashi,

I., (1998). The relationship between walking speed

and muscle strength for knee extension in hemiparetic

stroke patients: a follow-up study. The Tohoku

Journal of Experimental Medicine, 154(2), 111-113.

DOI: 10.1620/tjem.154.111

Olney, J. (1969). Brain lesions, obesity and other

disturbances in mice treated with monosodium

glutamate. Science, 164, 719-721.

DOI: 10.1126/science.164.3880.719

Park, S., Subramaniyam, M., Hong, S., & Kim, D. (2017).

Service Based Healthcare Monitoring System for the

Elderly - Physical Activity and Exercise, Proc.

International Conference on Applied Human Factors

and Ergonomics, Advances in Human Factors and

Ergonomics in Healthcare and Medical Devices,

337-342. Doi: 10.1007/978-3-319-60483-1_34

Patterson, K. K., Gage, W. H., Brooks, D., Black, S.

E., & McIlroy, W. E. (2010). Evaluation of gait

symmetry after stroke: a comparison of current

methods and recommendations for standardization.

Gait Posture, 31(2), 241-246.

Doi: 10.1016/j.gaitpost.2009.10.014.

Peter, G. A., & Arthur, D. K. (2016). Mechanisms of

Gait Asymmetry Due to Push-off Deficiency in

Unilateral Amputees. IEEE Transactions on Neural

Systems and Rehabilitation Engineering, 23(5),

776-785. DOI: 10.1109/TNSRE.2014.2356722

Roth, E. J., Merbitz, C., & Mroczek, K. (1997). Hemiplegic

gait. American Journal of Physical Medicine and

Rehabilitation, 76, 128-133.

DOI: 10.1097/00002060-199703000-00008

Turnbull, G. I., Charteris, J., & Wall, J. C. (1995). A

comparison of the range of walking speeds between

normal and hemiplegic subjects. Scandinavian Journal

of Rehabilitation Medicine Supplement, 27, 175-182.

Wall, J. C., & Turnbull, G. I. (1986). Gait asymmetires

in residual hemiplegia. Archives of Physical Medicine

and Rehabilitation, 67, 550-553.

Wevers, L., van de Port, I., Vermue, M., Mead, G., &

Kwakkel, G. (2009). Effects of task-oriented circuit

class training on walking competency after stroke:

a systematic review. Stroke, 40(7), 2450-2459.

DOI: 10.1161/STROKEAHA.108.541946

Yang, Y. R., Yen, J. G., Wang, R. Y., Yen, L. L., &

Lieu, F. K. (2005). Gait outcomes after additional

backward walking training in patients with stroke:

a randomized controlled trial. Clinical Rehabilitation,

19(3), 264-273. DOI: 10.1191/0269215505cr860oa

원고 수: 2019.06.19

수정 수: 2019.08.21

게재확정: 2019.08.21

ⓒ 2019 (by) the authors. This open access article is distributed under the terms and conditions of the Creative Commons Attribution license

(http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided that the original

work is properly cited.

gait feature vectors for post-stroke prediction using

Documents