Professional Documents
Culture Documents
Department of Orthopaedic Surgery, University of Heidelberg, Schlierbacher Landstr. 200a, 69118 Heidelberg, Germany
Orthopaedic Department, Papageorgiou Hospital, Thessaloniki, Greece
Behandlungszentrum Aschau, Germany
d
Institute for Applied Computer Science, Karlsruhe Institute of Technology (KIT), Germany
b
c
A R T I C L E I N F O
A B S T R A C T
Article history:
Received 27 July 2010
Received in revised form 13 January 2011
Accepted 15 January 2011
Hereditary spastic paraplegia (HSP) designates a group of genetic disorders typically leading to spasticity
in the lower limbs and consequently to gait disorders. Although the symptoms are similar to those of
cerebral palsy (CP), the correct diagnosis is important for treatment recommendations as one condition
is progressive in nature whereas the other is not. Due to the heterogeneity of HSP, genetic testing is
complex and in some genetic forms still not possible. The aim of this study was, therefore, to investigate
if instrumented 3D-gait analysis could help distinguish between these two conditions.
The gait pattern of 29 patients with HSP was compared with that of 29 patients with CP who were
matched in age, sex, and the extent of gait disturbance and also to 29 typically developing subjects for
reference. More than 3000 gait parameters were evaluated for their relevance to classify patients into
diagnostic groups. Cluster analysis revealed that these gait features may classify only subgroups of
symptoms as the gait pattern is very heterogeneous within each diagnosis group. However, prolonged
hip extension, knee extension, and ankle plantar exion were identied as indicators for HSP. In addition,
large trunk tilt velocities appear unique in some cases of HSP. These indicators in gait pattern may
contribute in establishing the diagnosis of HSP, which is important in predicting outcome when planning
surgical treatment for functional improvements in these patients.
2011 Elsevier B.V. All rights reserved.
Keywords:
Hereditary spastic paraplegia
Gait analysis
Automated methods
Classication
Feature extraction
Pattern recognition
1. Introduction
Hereditary spastic paraplegia (HSP) designates a heterogeneous
group of genetic disorders of which not all can yet be diagnosed by
routine genetic testing [1]. In its pure form, HSP leads to
progressive spasticity in the lower limbs due to pyramidal tract
degeneration [2]. HSP is classied as complex when complicated
by other neurological signs such as ataxia, mental retardation,
dementia, extrapyramidal signs, visual dysfunction, or epilepsy [3].
Pure HSP represents around 70% of the cases and is typically
inherited in an autosomal dominant manner [3,4]. In most cases,
the phenotype includes slowly progressive spasticity in the lower
limbs, resulting in gait deciencies and loss of mobility starting in
the second decade after the onset of symptoms. Clinically, the
onset of this type of HSP can be from childhood through to late
adult life, whereas more than half of mutation carriers do not
develop symptoms until after the age of 30 years [5]. Although the
clinical ndings may vary, two groups of clinical types have been
described based on the age at presentation: type I before and type II
after 35 years of age [6]. With an onset in the rst few years of life,
combined with delayed motor milestones, the diagnosis is more
suggestive of cerebral palsy (CP), particularly if the clinical picture
is relatively static [5]. However, due to the progressive nature of
the clinical symptoms in HSP, these patients are typically treated
differently to patients with CP. Since genetic testing is not always
possible and both the clinical status and the patients history may
not always be sufcient to establish the diagnosis, a detailed
analysis of the gait pattern may help in the decision for an
appropriate treatment.
Unlike CP, where gait has been studied in detail using 3D
motion capture, only few studies have objectively described gait
function in HSP. Braschinsky et al. [7] reported on reduced walking
speed in 46 patients set in the context of reduced active hip exion
and abduction ROM as well as reduced ankle dorsiexion ROM,
which were determined clinically. Earlier, the group of Klebe found
a reduced ROM in knee exion and reduced walking speed due to
reduced stride length and cadence along with an increase in step
width in 22 adult patients (age 3561 years) when compared to
557
Table 1
Subject characteristics.
Triple
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
HSP group
CP group
Reference
Sex
Age
GGI
Walking aid
Previous surgeries
HSP type
Sex
Age
GGI
Walking aid
Previous surgeries
Sex
Age
m
f
m
m
m
f
m
m
f
f
m
m
m
m
m
m
m
m
m
f
f
m
m
m
m
m
m
f
f
5.6
6.2
8.2
8.7
9.0
9.1
9.5
9.9
11.9
12.1
14.0
14.2
14.5
15.3
15.7
16.4
17.6
19.9
21.2
29.0
34.9
36.4
36.6
37.5
37.7
38.7
46.7
52.3
63.5
72
132
131
360
305
277
1511
45
606
271
100
118
149
545
994
398
299
1135
235
182
148
408
220
169
260
326
505
990
180
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
Yes
No
No
No
No
No
No
No
No
Yes
Yes
No
No
No
No
No
No
No
No
No
No
Yes
No
No
No
No
No
Yes
No
No
No
Yes
No
No
No
Yes
No
Yes
No
No
No
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
II
I
II
I
II
II
II
m
f
m
m
m
f
m
m
f
f
m
m
m
m
m
m
m
m
m
f
f
m
m
m
m
m
m
f
f
4.8
6.0
7.7
8.7
9.3
8.8
9.6
10.3
11.7
11.7
13.3
14.4
14.4
15.5
16.0
15.8
17.7
22.6
21.5
29.6
32.1
35.3
35.1
38.0
38.3
46.1
55.3
49.1
46.2
89
111
98
303
304
266
1584
137
581
279
128
189
235
526
1101
419
233
853
240
252
228
528
244
122
543
423
415
626
186
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
Yes
No
No
No
No
No
No
Yes
No
No
Yes
Yes
No
No
No
No
No
No
No
No
No
No
No
Yes
No
No
No
No
No
No
No
No
No
No
Yes
Yes
Yes
No
No
Yes
No
m
f
m
m
m
f
m
m
f
f
m
m
m
m
m
m
m
m
m
f
f
m
m
m
m
m
m
f
f
5.0
5.9
7.9
8.7
9.1
9.4
9.1
10.2
11.4
11.7
11.7
12.7
13.5
15.5
15.7
23.9
24.7
25.8
27.2
29.3
33.0
35.6
35.7
37.8
41.4
43.2
51.0
50.4
51.7
558
Table 2
List of the 10 most relevant gait parameters to distinguish the gait pattern of patients with HSP from the NORM using mutual information. TDP, time distance parameter; Str,
stride; St, stance; Sw, swing; MSw, mid swing.
Rank
Value
Time
series
Gait
phase
NORM
HSP
CP
Relevance
NORMHSP
p-Value
NORMHSP
Relevance
NORMCP
p-Value
NORMCP
1
2
3
4
5
6
7
8
9
10
ROM
ROM
TDP
ROM
ROM
ROM
MIN
ROM
TDP
ROM
Knee FlexExt
PelvicTilt
Double Support
Knee FlexExt
PelvicTilt Vel
Knee FlexExt
Knee FlexExt
PelvicTilt
Walking Speed
Knee FlexExt Vel
Sw
Str
54.2 (4.5)
2.8 (0.8)
20.7 (2.8)
29.4 (3.8)
0.4 (0.1)
55.5 (4.2)
3.9 (3.4)
2.5 (0.8)
1.3 (0.2)
6.6 (0.7)
26.1 (11.0)
8.9 (3.8)
31.6 (12.1)
10.0 (5.5)
1.2 (0.6)
38.4 (13.7)
21.4 (13.5)
8.1 (3.7)
0.9 (0.3)
4.1 (1.4)
23.4 (10.1)
9.3 (3.4)
33.5 (10.8)
8.0 (5.0)
1.1 (0.4)
34.9 (13.2)
27.7 (11.6)
8.5 (3.2)
0.8 (0.3)
3.8 (1.5)
0.587
0.580
0.519
0.485
0.476
0.456
0.454
0.454
0.453
0.445
1.89E
7.70E
1.19E
2.58E
3.49E
3.65E
2.36E
5.46E
4.78E
3.90E
0.628
0.745
0.561
0.540
0.524
0.500
0.603
0.606
0.560
0.476
2.72E
1.33E
1.98E
1.71E
7.64E
1.84E
8.39E
6.18E
7.91E
3.33E
MSw
Str
Str
Sw
St
Str
joint angles of pelvis, hip, knee, and ankle obtained by conventional, instrumented
3D gait analysis [11] and three derived time series, i.e., joint angle velocities, norm
distances, and norm distances for joint angle velocities to calculate single
features, i.e., scalar quantities of gait parameters. These parameters included range
of motion, mean, maximum, and minimum values of time series as well as the
temporal position of maximum and minimum values in the gait cycle or in
functional sub phases as dened by Perry [14]. In total, 3187 gait parameters,
including 19 additional scalar time distance parameters such as cadence and
walking speed, are generated. The feature set is very similar to [10], but without the
features from the stride-to-stride kinematic variance and with nine additional
time-distance parameters. To cover prolonged knee extension, the most successful
feature found in [9], the duration in the gait cycle with values smaller than zero (in
this case knee hyperextension) was systematically added as an additional
parameter for all joint angle and joint angle velocity time series. Driven by the
subjective clinical impression that trunk movement in some HSP patients appears
unique and distinct, 3D-data on trunk movements available in 18 (out of 29)
patients with HSP and 25 (out of 29) patients with CP were also analyzed via singlefeature computation in the same manner as described above.
The features were then ranked for relevance between zero (unable to separate
the different groups) and one (unique assignment) of the patient groups in question
using mutual information and a priori relevance. In other words, we assessed the
extent to which each parameter could suitably distinguish between gait patterns of
HSP and CP (and physiologic gait). Finally, when also taking cross correlations
between parameters into account by subsequently excluding variables when they
are less relevant than another highly correlated one, a ranking list with a set of
largely independent gait parameters is produced, characterizing the given clinical
(classication) problem. In this contribution, these lists were produced pairwise for
the groups HSP, CP, and NORM. Statistical signicance with a signicance level of
p < 0.05 was determined for each parameter using the non-parametric Wilcoxon
rank-sum test and Bonferroni correction for multiple testing.
Furthermore, unsupervised cluster analysis methods were applied to account for
heterogeneous gait patterns within each patient group. Here, a standard Fuzzy CMeans algorithm with raw data-based time series, a normalization of time series to
mean values of zero and a standard deviation of one, a fuzzier m = 1, and a xed
number of clusters C = 4, 5, and 6 were chosen [15]. Sagittal plane kinematics
including joint angles and joint angle velocities of pelvis, hip, knee, and ankle was
used to classify the sets of 4, 5, and 6 clusters across patient groups. The analysis was
done with the open-source Matlab toolbox Gait-CAD [19,20].
4. Results
In Table 2, a ranking list is given for the 10 most relevant gait
parameters classifying the group of patients with HSP from agematched reference normal subjects (NORM), corresponding values
19
16
13
20
14
11
15
15
13
16
20
19
17
20
19
14
20
20
18
17
Table 3
List of the 10 most relevant gait parameters distinguishing the gait patterns of patients with HSP from those with CP. Values: TDP, time distance parameter; T < 0 = time
duration in which the parameter is smaller then 0; ROM, range of motion. Gait phases: Str, stride; St, stance; Sw, swing; TSt, terminal stance.
Rank
Value
Time series
1
2
3
4
5
6
7
8
9
10
TDP
TDP
Mean
T<0
T<0
T<0
Mean
ROM
MIN
Mean
Stride Time
Step Time
PelvicTilt
Hip FlexExt
Hip FlexExt
DorsiPlanFlex
PelvicTilt
Knee Rotation
Pelvic Rotation
Hip FlexExt
Gait phase
NORM
HSP
CP
Relevance HCPCP
p-Value HSPCP
Str
St
Str
Str
St
Str
Sw
TSt
1.0 (0.1)
0.6 (0.1)
11.1 (4.3)
20.0 (8.6)
22.6 (10.6)
39.9 (11.9)
11.2 (4.3)
17.2 (4.6)
4.5 (2.5)
0.7 (5.6)
1.3 (0.7)
0.7 (0.3)
17.5 (6.7)
12.7 (11.9)
13.7 (13.3)
55.5 (36.3)
17.4 (6.6)
14.5 (5.2)
7.3 (5.9)
7.2 (11.2)
1.2 (0.4)
0.6 (0.2)
16.8 (7.3)
6.4 (10.7)
7.1 (12.5)
34.9 (32.7)
16.7 (7.4)
11.8 (5.5)
7.4 (9.6)
11.8 (9.9)
0.105
0.104
0.082
0.080
0.077
0.073
0.067
0.061
0.060
0.060
0.793
0.529
0.722
0.003
0.003
0.002
0.657
0.002
0.976
0.004
PelvicTilt
Crouch
40
20
20
HipFlexExt
DorsiPlanFlex KneeFlexExt
80
60
40
20
0
50
50
100
100
50
50
100
100
80
60
40
20
0
0
50
100
20
0
-20
-40
-60
-80
50
100
100
50
100
50
50
100
100
80
60
40
20
0
0
50
100
20
0
-20
-40
-60
-80
0
50
100
Gait cycle [%]
0
0
60
40
20
0
-20
60
40
20
0
-20
50
100
50
100
50
100
80
60
40
20
0
0
50
100
20
0
-20
-40
-60
-80
0
50
100
Gait cycle [%]
Norm like
40
20
0
50
80
60
40
20
0
20
0
-20
-40
-60
-80
0
50
100
Gait cycle [%]
20
0
60
40
20
0
-20
Jump Knee
40
0
0
60
40
20
0
-20
Stiff Knee
40
20
0
0
60
40
20
0
-20
Recurvatum
40
559
20
0
-20
-40
-60
-80
0
50
100
Gait cycle [%]
0
50
100
Gait cycle [%]
Fig. 1. Clusters of sagittal gait kinematics found across all subjects (HSP, CP, and NORM).
Jump Knee
Jump Knee
number of legs
45
40
35
Stiff Knee
30
Stiff Knee
25
20
Norm like
Recurvatum
15
10
5
0
Crouch
Crouch
Norm like
Norm like
CP
HSP
Norm
Fig. 2. Relative distribution of each patient group into gait pattern clusters.
560
561
[15] Rozumalski A, Schwartz MH. Crouch gait patterns dened using k-means
cluster analysis are related to underlying clinical pathology. Gait Posture
2009;30(2):15560.
[16] Sutherland DH, Davids JR. Common gait abnormalities of the knee in cerebral
palsy. Clin Orthop Relat Res 1993;288:13947.
[17] von Fellenberg J, Paternotte C, Prudhomme JF, Weissenbach J, Hazan J,
Burgunder JM. Klinische und molekulargenetische Analyse bei vier schweitzerischen Familien mit der reinen Form der hereditaren spastischen Spinalparalyse. Schweiz Med Wochenschr 1998;128(26):104350.
[18] Erichsen AK, Koht J, Stray-Pedersen A, Abdelnoor M, Tallaksen CM. Prevalence
of hereditary ataxia and spastic paraplegia in southeast Norway: a populationbased study. Brain 2009;132(Pt 6):157788.
[19] Mikut R, Burmeister O, Braun S, Reischl M. The open source Matlab toolbox
Gait-CAD and its application to bioelectric signal processing. In: Proceedings of
DGBMT-Workshop Biosignal Analysis, Potsdam; 2008. p. 10911.
[20] http://sourceforge.net/projects/gait-cad/.