32nd Annual International Conference of the IEEE EMBS Buenos Aires, Argentina, August 31 - September 4, 2010
Gait initiation evaluation after deep brain stimulation for Parkinson’s disease: a 7-year follow-up A. M. S. Muniz, Member EMBS, W. Liu, H. Liu, K. E. Lyons, R. Pahwa, and J. Nadal, Member, IEEE
Abstract—This study evaluated the long-term effects of deep brain stimulation of the subthalamic nucleus (DBS-STN) on gait initiation. Six Parkinson’s disease (PD) patients who had undergone DBS-STN and 31 control subjects were evaluated. PD subjects were assessed at two different time periods: 11.3 ± 10.3 (P1) and 78.9 ± 10.6 (P2) months after surgery. Subjects under stimulation were tested in two conditions: without medication and with medication. Principal components (PC) analysis was separately applied on vertical, anterior-posterior and medial-lateral ground reaction force (GRF) from gait initiation, during the anticipatory postural adjustment (APA) phase. Three PC scores were chosen by the scree test for each GRF component. The higher loading factors pointed to major differences between controls and PD patients on maximum APA amplitude for vertical and anterior-posterior GRF. Friedman test showed a significant difference in standard distance among conditions (P = 0.006), with the post-hoc test recognizing differences only between P1 and P2 in the medication-on condition. All distances increased in the follow-up evaluation (P2), when considering the same medication condition, indicating a worsening in gait initiation after 7 years of follow-up.
I. INTRODUCTION
P
arkinson’s disease (PD) is a progressive, neurodegenerative disorder affecting dopaminergic neuronal systems, with a consequently impaired motor function. Bradykinesia, tremor, rigidity and postural instability are the typical motor symptoms of PD [1]. Loss of dopaminergic neurons leads to a series of changes in the electrical activity of basal ganglia structures [2]. These changes result in a short-stepped and shuffling walking pattern that progressively affects mobility and independence [3]. Impaired gait initiation in PD patients is a typical functional sign of akinesia, a failure or slowness of willed movement. Gait initiation is defined as the transient state between standing and walking, being an important diagnostic
Manuscript received April 01, 2010. This work was supported in part by the Brazilian Research Council (CNPq). A. M. S. Muniz is with Department of Post-graduation, Physical Education Collage of Brazilian Army, Rio de Janeiro, RJ – BRAZIL, (email:
[email protected]) H. Liu and W. Liu are with Department of Physical Therapy and Rehabilitation Sciences, University of Kansas Medical Center, Kansas City, KS, USA (emails:
[email protected],
[email protected]) K. E. Lyons and R. Pahwa are with the Parkinson’s Disease and Movement Disorder Center, Department of Neurology, University of Kansas Medical Center, Kansas City, KS, USA (emails:
[email protected],
[email protected]). J. Nadal is with the Biomedical Engineering Program, Federal University of Rio de Janeiro, Brazil - Rio de Janeiro, RJ – BRAZIL (
[email protected])
978-1-4244-4124-2/10/$25.00 ©2010 IEEE
tool to study pathologic gait. In the postural phase, which corresponds to the generation of propulsive forces [4], the center of pressure (COP) is first moved posteriorly and towards the swing foot that takes the first step, followed by a weight shift towards the supporting foot as the swing foot begins to leave the ground. The increased duration of the postural phase, decreased propulsive forces and reduced lateral shift of body mass over the stance limb are the main features of gait initiation disorder in PD patients. The PD pharmacological treatment has been able to provide adequate symptomatic control in the first 5-10 years of therapy [5]. In the last decade, surgical procedures have been increasingly used in advanced PD stages due to the introduction of high-frequency stimulation [6]. Deep brain stimulation of the subthalamic nucleus (DBS-STN) relieves motor symptoms and allows medication reduction [7]. The effects of DBS-STN on gait initiation include shortening of the imbalance phase, larger backward/lateral displacement of COP and more physiological expression of the underlying anticipatory muscular synergy [8]. Liu et al. [9] found a significant increase in the amplitude of the anticipatory postural adjustment (APA) phase, with increased amplitudes of reactive shear forces on both feet and COP displacements. In previous studies, principal component analysis (PCA) was applied to ground reaction forces (GRF) during gait to classify normal and pathologic patterns, as well as to assess treatment effects [10]-[11]. PCA allows obtaining a reduced set of uncorrelated parameter, which takes into account the whole GRF signal [12]. Long-term clinical effects of DBS-STN have been described [5], [13] - [16]; however, the long-term DBS-STN outcomes on gait, specifically gait initiation, have not been reported. Consequently, the aim of this study is to evaluate the long-term effects of DBS-STN on gait initiation. II. MATERIALS AND METHODS A. Sample This study analyzed data from six PD patients (two women); average age of 50.4 ± 7.87 (mean ± standard deviation) who had undergone bilateral STN DBS and were stable when the study was conducted. A control group with 31 healthy subjects (20 women) average age of 50.1 ± 7.8 years who had no history of neurological illness, degenerative conditions or any other disease that might interfere with body
3650
B. Testing procedure PD patients were evaluated at two different time periods 11.3 ± 10.3 (P1) months after surgery and 78.9 ± 10.6 (P2) months after-surgery. Duration of PD was 13.7 ± 4.59 years at P1. In each experiment, subjects came to the laboratory on two different days for repeated quantitative gait initiation measurements. At the P1 period, before coming to the research laboratory in the first visit, the subject had taken the usual dose of PD medications and stimulators were turned “on”. Gait initiation assessment was first conducted with both medication and stimulation (P1mon-son condition). After turning the stimulator off for 30 min, the measurements were repeated (P1mon-sof). In the second visit, the subjects were without medication for at least 12 hours. Gait initiation analysis was first conducted with stimulation (P1mof-son), and repeated after 30 minutes without stimulation (P1mof-sof). At the P2 period, the experiments were conducted with stimulation and medications in the first visit (P2mon-son), and then without medications in the second visit (P2mof-son), since it was very difficult for the subjects to sustain stability without stimulation. One subject was not taking any PD medications (subject #6), and his data was only analyzed in the P2mof-son condition. Subjects from the control group were evaluated only once. Two force platforms (AMTI, Advanced Mechanical Technology, USA) were used to record the gait initiation data. Four trials of gait initiation tests were conducted under each medication/stimulation condition. The first trial was not included to avoid bias due to the familiarization to each test condition, while the second to fourth trials were used in analysis. During each trial, the subjects stood barefoot with one foot on each force platform, arms at their sides, and looking straight ahead at a red lamp (visual triggering signal) located 4 m ahead at their eye level. The subjects were instructed to start walking forward just after the light turned on at a self-selected speed. Data acquisition began approximately 1 s before the light trigger, and ended about 1 s after the first heel strike. The GRF signals from the APA were collected at a frequency of 100 Hz. C. Signal Processing The data were filtered using a low pass Butterworth filter, nd 2 order with a cut-off frequency 30 Hz, applied in direct and reverse order to avoid phase distortion, and normalized by the subject’s body weight. Force platform measures from gait initiation included the averaged vertical, anterior-posterior and medial-lateral GRF under the swing and stance limb (Fig. 1). The averaged components of GRF from three trials during the APA phase of gait initiation were interpolated with cubic
splines and re-sampled to have 100 sample points for each foot. Thus, 200 GRF samples, that corresponded to swing and stance legs were separately analyzed for each GRF component. Each vertical, anterior–posterior and medial–lateral GRF waveform was stored in a matrix E (37 x 200). Each row corresponded to one subject, including 31 controls and six PD patients in the P1mof-son condition, which was a priori assumed as the reference condition for further comparisons. Each column contained the GRF samples. PCA was applied to the covariance matrices S (200 x 200) from each E, in the following linear system [12]:
Sx p = λ p x p
(1)
where p are the eigenvalues organized in decreasing order and xp are the corresponding eigenvectors or principal components (PCs). These PCs are independent waveform features based on the original waveform variability. The firsts PC correspond to the largest sources of variation, being orthogonal to each other. For each GRF component, the relevant PCs for the analysis were selected by the scree test [12]. Each value of a PC represents the loading factor applied to the corresponding sample of the original waveforms [17]. The signal from each subject is represented by the PC scores, given by the scalar product between the original GRF samples, after removing the ensemble mean (m) obtained in the S calculation, and the respective PC [12]. All signal processes were implemented in Matlab 6.5 (The Mathworks, USA) a) 1.2 0.6
% Body weight
sway or gait was also analyzed. Each subject signed an informed consent approved by the Institutional Review Board of the University of Kansas Medical Center.
0 0 b) 0.1
20
40
60
80
100
20
40
60
80
100
40
60
80
100
0 -0.1 0 c) 0.05 0 -0.05 0
20
% of gait initiation cycle
Fig 1. Example of APA from gait initiation of a normal subject on swing (continuous) and stance (dotted) leg: a) Vertical GRF; b) Anterior-posterior GRF; and c) Medial-lateral GRF.
D. Standard Distance The standard distance proposed by Flury and Riedwyl [18] was calculated including the selected PC from each GRF component. This index represents the distance between each PD observation and the center of the ellipsis of control subjects. For classifying normal or abnormal APA from gait initiation, the cut-off point between standard distance values from the control group and PD subjects in the P1mof-son condition was obtained by logistic regression [10]. The classifier performance was assessed by the leave-one-out cross-validation technique, which provides a good indication
3651
% Body weight
0.2
0 0
50
100
150
200
50
100
150
200
0
-0.2 0
Fig. 2. a) Averaged vertical GRF from control group (continuous) and PD in P1mof-son condition (dashed); and b) The corresponding first principal component (The arrows indicate the main loading factors). In x-axis, points between 0 and 100 correspond to the swing limb and between 101 and 200 correspond to the stance limb. a)
0.05
-0.05
0
-0.1 -0.15 0
III. RESULTS
b)
0.05
-0.05
50
100
150
200
50
100
150
200
0
-0.1 -0.15 0
Fig. 3. a) Averaged anterior-posterior GRF from control group (continuous) and PD in mof-son condition (dashed); and b) The corresponding first principal component (The arrows indicate the main loading factors). In x-axis, points between 0 and 100 correspond to the swing limb and between 101 and 200 correspond to the stance limb. TABLE I
STANDARD DISTANCE OF PD IN THE FOUR DIFFERENT CONDITIONS Subjects 01 02 03 04 05 06 Mean ± SD*
P1mof-son
P1mon-son
P2mof-son
P2mon-son
17.10 18.01 16.90 23.21 16.58 6.48 16.38 ± 5.44
16.57 18.47 16.14 19.68 9.84 13.36 15.68 ± 3.59
39.16 30.25 32.34 26.96 20.71 7.42 26.14 ± 11.00
25.46 30.56 26.01 25.64 33.89 N.A. 28.31 ± 3.77
*Significant difference among conditions (Friedman test, P = 0.006), specifically between P1mon-son and P2mon-son conditions (Post-hoc Dunn test, P < 0.05). Bold values are classified as normal. The cut-off point given by logistic regression was 15.8. 40
35
30
Standard Distance
According to the scree test, three PCs from each GRF component were attained for analysis, representing respectively, 93.78%, 90.02% and 89.58% of total variance of each signal. Only the first PC scores from vertical (p = 0.0028) and anterior-posterior (p = 0.0199) GRF were statistically different between controls and PD subjects in the P1mof-son condition. The first PC presented higher loading factors during the APA peaks in both swing and stance limbs (arrows in Fig. 2b), presenting a waveform similar to the vertical GRF (Fig. 2a). The anterior-posterior GRF presented higher loading factors at 60-70% of swing phase for the swing limb and at the end for the stance limb (arrows in Fig. 3). The other eigenvectors were not presented, since they did not show statistical difference between controls and PD patients. The standard distance allowed classifying control and PD subjects in the P1mof-son condition with accuracy 0.94, sensitivity 0.83 and specificity 0.97. The cut-off point determined by logistic regression was 15.8, with lower values being classified as normal APA. Friedman test showed a significant difference in standard distance among conditions (P = 0.006), with the post-hoc test recognizing differences only between P1mon-son and P2mon-son conditions (Table I). It should be noted that all distances increased in follow-up evaluation P2, when considering the same medication condition (Table I, Fig. 4). Removing the medication on each evaluation stage tends to increase distances as shown by the medians in Fig. 4, however without statistical significance.
0.5
% Body weight
F. Statistical Analysis Wilcoxon rank sum test was applied to verify statistical differences in PC scores between controls and PD subjects. Non-parametric Friedman test was applied to verify differences among PD treatments in the standard distance. Post-hoc analysis was performed with the Dunn test. The significance level was . = 0.05. All statistical tests were performed in SPSS 17.0 (SPSS, USA).
b)
1.1
Loading Factor
E. Follow-up evaluation The PC scores from PD subjects in the follow-up condition (P1mon-son, P2mof-son, P2mon-son) were obtained from each condition. Standard distance values were also calculated for these conditions.
a)
Loading Factor
of reliability in classification of small datasets. The results were used to performance evaluation by computing overall accuracy, sensitivity (correct classification of PD) and specificity (correct classification of controls).
25
20
15
10
5
controls
P1mon-son P1mof-son
P2mon-son P2mof-son
Fig. 4. Boxplot of standard distance from normals and PD subjects. The box indicate the 1st and 3rd quartiles, the internal line the median, and “+” represent
3652
outliers.
IV. DISCUSSION This is the first study focused on measures prior to foot lifting during gait initiation (APA phase) for assessing the long-term effects of DBS-STN. The surgery greatly reduces Parkinsonian motor symptoms and drug-induced dyskinesia, and improves patients’ ability to perform activities of daily living, like walking [7] with fewer motor fluctuations. Liu et al. [9] and Crenna et al. [8] found improvements in the gait initiation process in PD patients who underwent DBS-STN. However, only discrete parameters were evaluated, as the maximum amplitudes and times. According to Tingle et al. [19], such approach neither considers the high degree of correlation that exists between various aspects of an individual’s gait variable nor includes the information that may lie in the pattern of the waveform. Conversely, PCA coefficients account information from the entire GRF waveforms, providing a substantial dimension reduction [12]. The analysis of eigenvectors may shed light on the interpretation of PC analysis. Each PC sample constitutes the loading factor attributed to the corresponding sample of the original GRF, with higher absolute loading factors pointing to the epochs of higher variance between groups within the original waveforms [18]. The comparison between the first PC and the original GRF shapes point to differences between groups mainly around the peaks of APA on vertical GRF (Fig. 2), and maximum amplitude of swing leg and at the end of stance leg on the anterior-posterior GRF (Fig. 3). Liu et al. [9] described that the amplitude of APA on vertical and anterior-posterior were good indicators of the DBS effect in improving PD patients’ gait initiation. When comparing visually the waveforms of controls and PD (Fig. 2a and Fig. 3a) the PD patients presented smaller GRF amplitudes. The amplitude of APA is related to the propulsive force generated on the swing foot before the swing foot lifting [20]. The standard distance calculated by PC scores from vertical, anterior-posterior and medial-lateral GRF in APA phase evidenced a good classifier performance, similar to what was found using vertical GRF in gait analysis [10]. The smallest values were found in the P1mon-son condition, with two subjects classified as normal. By comparing the loading factor analysis (Fig 2b and Fig 3b), this result means that in P1mon-son conditions, PD subjects presented APA amplitudes closer to normal pattern. Crenna et al. [8] found larger backward/lateral displacement of COP and more physiological expression of the underlying anticipatory muscular synergy with DBS-STN. Liu et al. [9] found the highest amplitudes in the mon-son condition. All standard distances increased in the P2 condition either with or without medication (Table I and Fig. 4), which means a worsening in gait initiation after 7-year follow-up. This
result agrees with the analysis of clinical scores [5], [15]-[16], that showed the beneficial effects of DBS-STN compared to pre-surgery conditions persisted after a follow-up period of 4-5 years, particularly in the “off” medication condition; however, such improvement was lower than ones obtained 1-2 years after surgery. These results likely reflect disease progression rather than a side effect of DBS-STN [13]. Results pointed to the potential power of standard distance for classifying long-term PD treatment effects derived from APA gait initiation. The eigenvector analysis evidenced the most difference between groups in the maximum APA amplitude for vertical and anterior-posterior GRF. Moreover, the standard distance increased in the 7-year follow-up evaluation of DBS-STN, when considering the same medication condition, indicating a worsening in gait initiation. REFERENCES [1]
[2] [3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12] [13]
[14]
[15]
3653
M. E. Morris, F. Huxham, J. Mcginley et al., “The biomechanics and motor control of gait in Parkinson disease,” Clin Biomech, vol.16, pp. 459-470, 2001 A. M. Lozano, “Deep brain stimulation for Parkinson’s disease,” Parkinsonism Relat Disord, vol. 7, pp.199-203, 2002. D. M. Urquhart, M E. Morris, R. Iansek, “Gait consistency over a 7-day interval in people with Parkinson’s disease,” Arch Phys Med Rehabil, vol. 80, pp.696-701, 1999. C. Crenna, C. Frigo, “A motor programme for the initiation of forward-oriented movements in humans,” Journal Physiol vol. 437, pp.635–653, 1991. M.C. Rodriguez-Oroz, J. A. Obeso, A. E. Lang, et al., “Bilateral deep brain stimulation in Parkinson’s disease: a multicenter study with 4 years follow-up,” Brain, vol. 128, pp.2240-2249, 2005. S. Lubik, W. Fogel, V. Tronnier, et al., “Gait analysis in patients with advanced Parkinson disease: different or additive effects on gait induced by levodopa and chronic STN stimulation,” J Neural Transm, vol.113, pp.163-173, 2006. N. Allert, J. Volkmann, S. Dotse, et al., “Effects of bilateral pallidal or subthalamic stimulation on gait in advanced Parkinson’s disease,” Mov Disor, vol. 16, no.6, pp.1076-1085, 2001. P. Crena, I. Carpinella, M. Rabuffetti, et al., “Impact of subthalamic nucleus stimulation on the initiation of gait in Parkinson’s disease,” Exp. Brain Res, vol.172, pp.519-532, 2006. W. Liu, K. McIntire, S.H. Kim, et al., “Bilateral subthalamic stimulation improves gait initiation in patients with Parkinson’s disease,” Gait Posture, vol. 23, pp.292-498, 2006. A. M. S. Muniz, J. Nadal, “Application of principal component analysis in vertical ground reaction force to discriminate normal and abnormal gait,”Gait Posture, vol.29, pp.31-35, 2009. A. M. S. Muniz, H. Liu, K.E. Lyons, R. Pahwa, W. Liu, F.F. Nobre, J. Nadal, “Comparison among probabilistic neural network, support vector machine and logistic regression for evaluating the effect of subthalamic stimulation in Parkinson disease on ground reaction force during gait,” J Biomech vol.43, pp.720-726, 2010. I. T. Jolliffe, “Principal Component Analysis”. 2nd ed. New York: Springer, 2002. V. Visser-Vandewalle, C. Linden, Y. Temel, Y., et al., “Long-term effects of bilateral subthalamic nucleus stimulation in advanced Parkinson disease: a four year follow-up study,” Parkinsonism Relat Disord vol.11, pp.157-165, 2005. S. Sestini, S. Ramat, A. R. Formiconi, et al., “Brain networks underlying the clinical effects of long-term subthalamic stimulation for Parkinson’s disease: a 4-year follow-up study with rCBF SPECT,” The Journal of Nuclear Medicine vol. 46, no.9, pp.1444-1454, 2005. K. Ostergaard, N. A. Sundae, “Evolution of Parkinson’s disease during 4 years of bilateral deep brain stimulation of the subthalamic nucleus,” Mov. Disor., vol. 21, no.5, pp.624-631, 2006.
