- Email: [email protected]
Analysis of tremor in multiple sclerosis using Hilbert-Huang Transform
+Model NEUCLI-2502; No. of Pages 10
ARTICLE IN PRESS
Neurophysiologie Clinique/Clinical Neurophysiology (2015) xxx, xxx—xxx
Disponible en ligne sur
ScienceDirect www.sciencedirect.com
ORIGINAL ARTICLE/ARTICLE ORIGINAL
Analysis of tremor in multiple sclerosis using Hilbert-Huang Transform Analyse du tremblement de la sclérose en plaques utilisant la transformée de Hilbert-Huang S.-S. Ayache a,b, T. Al-ani a, W.-H. Farhat a,b, H.-G. Zouari a,b,c, A. Créange a,d, J.-P. Lefaucheur a,b,∗ a
EA4391, faculté de médecine de Créteil, université Paris Est—Créteil, 8, avenue du Général-Sarrail, 94010 Créteil cedex, France b Service de physiologie et d’explorations fonctionnelles, hôpital Henri-Mondor, Assistance publique—Hôpitaux de Paris, 51, avenue du Maréchal-de-Lattre-de-Tassigny, 94010 Créteil cedex, France c Service d’explorations fonctionnelles, CHU Habib Bourguiba, Sfax, Tunisia d Service de neurologie, hôpital Henri-Mondor, Assistance publique—Hôpitaux de Paris, 51, avenue du Maréchal-de-Lattre-de-Tassigny, 94010 Créteil cedex, France Received 15 February 2015; accepted 27 September 2015
KEYWORDS Accelerometer; Action tremor; Electromyography; Empirical mode decomposition; Multiple sclerosis; Signal analysis
∗
Summary Tremor is frequently described in patients with multiple sclerosis (MS) but remains poorly characterized using neurophysiological techniques. Accelerometric (ACC) and electromyographic (EMG) recordings were performed in 26 MS patients complaining of clumsiness, associated (n = 16) or not associated (n = 10) with visible tremor. Seventeen healthy subjects with physiological tremor (PT) and eight patients with essential tremor (ET) served as controls. Signals were analyzed using non-linear Empirical Mode Decomposition (EMD) and related HilbertHuang Transform (HHT), compared to the standard linear spectral analysis using Fast Fourier Transform (FFT). The presence of cerebellar signs and motor deficit was assessed on clinical examination. Using FFT, tremor was found in all patients with ET and 12% of subjects with PT, but in none of the MS patients, even in the presence of visible tremor. In contrast, EMD-HHT analysis of ACC-EMG coupling showed common frequency peaks characterizing tremor related to a central generator in 62.5% of MS patients with visible tremor, 40% of MS patients without visible tremor, 29% of subjects with PT, and all patients with ET. In EMD-HHT analysis, tremor characteristics were similar in subjects with PT and MS patients, regardless of the presence of a visible tremor, but these characteristics clearly differed in patients with ET. A visible tremor in MS patients was associated with more frequent cerebellar signs and less motor deficit at
Corresponding author. E-mail address: [email protected] (J.-P. Lefaucheur).
http://dx.doi.org/10.1016/j.neucli.2015.09.013 0987-7053/© 2015 Elsevier Masson SAS. All rights reserved.
Please cite this article in press as: Ayache S-S, et al. Analysis of tremor in multiple sclerosis using Hilbert-Huang Transform. Neurophysiologie Clinique/Clinical Neurophysiology (2015), http://dx.doi.org/10.1016/j.neucli.2015.09.013
+Model NEUCLI-2502; No. of Pages 10
ARTICLE IN PRESS
2
S.-S. Ayache et al. the upper limb. The low-frequency tremor observed in MS patients could therefore originate in lesions of the brainstem (midbrain) or cerebellothalamic circuits, or may correspond to an enhanced PT, partly favored by cerebellar dysfunction and being more visible during movement execution in the absence of concomitant motor deficit. © 2015 Elsevier Masson SAS. All rights reserved.
MOTS CLÉS Accéléromètre ; Analyse du signal ; Décomposition en mode empirique ; Électromyographie ; Sclérose en plaques ; Tremblement d’action
Résumé Le tremblement est souvent décrit chez les patients atteints de sclérose en plaques (SEP), mais reste mal caractérisé sur le plan neurophysiologique. Des enregistrements accélérométriques (ACC) et électromyographiques (EMG) ont été réalisés chez 26 patients atteints de SEP qui se plaignaient de maladresse, associée (n = 16) ou non (n = 10) à un tremblement visible cliniquement. Dix-sept sujets sains présentant un tremblement physiologique (TP) et huit patients atteints de tremblement essentiel (TE) ont servi de témoins. Les signaux ont été analysés par une méthode non-linéaire de décomposition en mode empirique (EMD) associée à une transformée de Hilbert-Huang (HHT), comparée à l’analyse spectrale linéaire classique, utilisant une transformée rapide de Fourier (FFT). La présence de signes cérébelleux et d’un déficit moteur a été évaluée cliniquement. En utilisant la FFT, un tremblement a été trouvé chez tous les patients avec TE et 12 % des sujets avec TP, mais chez aucun patient atteint de SEP, même en présence d’un tremblement visible cliniquement. En revanche, l’analyse EMD-HHT du couplage ACC-EMG a montré des pics de fréquence communs témoignant d’un tremblement avec générateur central chez 62,5 % des patients atteints de SEP avec tremblement visible, 40 % des patients atteints de SEP sans tremblement visible, 29 % des sujets avec TP, et chez tous les patients avec TE. Dans cette analyse EMD-HHT, les caractéristiques du tremblement étaient similaires entre les sujets avec TP et les patients atteints de SEP, indépendamment de la présence d’un tremblement visible, mais il est clair que ces caractéristiques différaient chez les patients avec TE. Le tremblement visible chez les patients atteints de SEP était associé à des signes cérébelleux plus fréquents et à un moindre déficit moteur. Ainsi, le tremblement de basse fréquence observé chez les patients atteints de SEP pourrait avoir pour origine une lésion du tronc cérébral (mésencéphale) ou de circuits cérébellothalamiques ou bien correspondre à un TP exagéré, en partie favorisé par une dysfonction cérébelleuse et que l’absence de déficit moteur concomitant rendrait plus visible cliniquement pendant l’exécution d’un mouvement. © 2015 Elsevier Masson SAS. Tous droits réservés.
Introduction Tremor was found to be clinically present in 25% and 58% of patients with multiple sclerosis (MS) in two previously published large cohorts [2,40]. In MS patients, tremor usually consists of an action tremor of the upper extremities, the rest component being almost inexistent [2]. However, MS patients may have various types of action tremor intermingled with other movement disorders in a complex clinical picture [1,2,28]. It may be difficult to characterize tremor in this context and neurophysiological investigations can be helpful. Clinical neurophysiology of tremor is usually based on the analysis of accelerometric (ACC) and electromyographic (EMG) recordings with Fast Fourier Transform (FFT), including the study of peaks in the power spectral analysis of periodograms, regularity in the time-frequency domain, asymmetry of the autocorrelation function decay, or coupling by applying cross-correlation and coherence analyses [11,21,36,38]. Despite the contribution of FFT methods in this domain, shown in these and other works, the appropriateness of using these methods is still limited since it is based on the assumption that tremor dynamics reflect periodic and linear activity, which is in fact not true [39,48]. Indeed tremor is a non-periodic, non-linear and non-stationary
signal with time-varying frequency, leaving room for nonlinear methods of signal analysis. Another limitation of FFT application is that it cannot distinguish between different types of tremor, whose frequency overlaps in similar ranges. In this context, Empirical Mode Decomposition (EMD) [26] and related Hilbert-Huang Transform (HHT) [24] could be relevant, according to their ability to deal with a nonstationary signal and their capacity to detect low-frequency components embedded in a non-stationary background [31]. In a previous pilot study including healthy subjects and patients with essential tremor (ET) [3], we described various EMD-HHT procedures for tremor analysis. The objective of the present study was to determine whether these procedures could be applied in MS patients to better characterize tremor in this particular clinical condition. In addition, tremor characteristics observed in MS patients were compared to those of other types of tremor (physiological tremor [PT] and ET), using the same signal analysis approach.
Methods Participants Sixteen MS patients with visible tremor in one or both upper extremities and ten MS patients without visible tremor were
Please cite this article in press as: Ayache S-S, et al. Analysis of tremor in multiple sclerosis using Hilbert-Huang Transform. Neurophysiologie Clinique/Clinical Neurophysiology (2015), http://dx.doi.org/10.1016/j.neucli.2015.09.013
+Model NEUCLI-2502; No. of Pages 10
ARTICLE IN PRESS
EMD-HHT analysis of MS tremor enrolled from the Department of Neurology of Henri Mondor hospital. All these patients complained of clumsiness in one or both upper limbs. Clumsiness was defined as the incapacity to perform at least one of the daily life activities listed in the Chedoke Arm and Hand Activity Inventory [4]. Patients were also selected according to the following criteria: • confirmed diagnosis of MS in reference to the revised McDonald’s criteria [41]; • no history of MS relapse in the previous month; • age between 18 and 70 years; • absence of any other neurological or psychiatric disease; • signed informed consent. Two groups of patients served as controls: a first group of 17 healthy subjects with PT and a second group of eight patients with ET. PT and ET were defined according to the consensus statement of Movement Disorder Society on tremor [13].
Clinical neurological examination MS patients underwent standard neurological examination with special attention paid to the most affected upper limb, to determine the presence or absence of a visible tremor, cerebellar signs (dysdiadochokinesis and dysmetria), and motor deficit. For the latter purpose, we calculated a motor score ranging from 0 (null strength) to 20 (full strength), as the summation of the scores measured for four muscle groups (pinch, wrist extension, forearm flexion and arm abduction) using the Medical Research Council (MRC) scale for muscle strength [33]. In addition, the Expanded Disability Status Scale (EDSS) was scored for each MS patient [30].
Neurophysiological investigation The same investigator (SSA) performed neurophysiological investigation, consisting of ACC and EMG recordings. Patients and subjects were seated in a comfortable chair with the upper limb fully at rest or held in the ‘‘oath position’’. Tremor evaluation was carried out on the side most severely affected by tremor or clumsiness. If both sides were equally affected, we chose to evaluate the right one. For ACC recordings, a piezoelectric single plane accelerometer (TREM0000, Neuroservices, Evry-Lisses, France) was fixed on the dorsal aspect of the distal phalanx of the index finger. Bipolar EMG was recorded from the extensor and flexor carpi radialis muscles, using a pair of surface pre-gelled electrodes (Ref 9013S0242, NatusDantec, Skovlunde, Danemark), placed 2 cm apart over the muscle belly (see the picture of the experimental setup in [3]). Thus, both ACC and EMG recordings concerned the same movement (extension/flexion of the hand and fingers). All recordings were performed with a Keypoint machine (Natus-Dantec). Signals were amplified, filtered (band-pass: 1—20 Hz for ACC signals; 20—1000 Hz for EMG signals) and uniformly sampled at 24 kHz. To shorten the computation time, the ACC and EMG data were downsampled to 1000 Hz using a polyphase implementation in Matlab software (MathWorks, Natick, MA, USA). The resampled signals were then filtered over the (0—15 Hz) frequency range. This offline
3 analysis allowed study of the envelope of the tremor signal, according to the method developed by Fox and Randall [20], then used, for example, by Elble and Randall [17,18] or O’Suilleabhain and Matsumoto [38]. Prior to frequency analysis, these authors low-pass filtered at 20—21 Hz the raw EMG signal recorded with a band-pass of 10—3000 Hz or 30—500 Hz. The resulting envelope of the surface EMG provides a measure of any rhythmical variations within the frequency band of tremor [17,18]. Similarly, in the present study, we looked at the tremor envelope of the EMG signal and not at the pattern of motor unit recruitment during voluntary muscle contraction, whose frequency analysis requires higher and broader band-pass. Actually, during voluntary muscle contraction, it is known that single motor unit action potentials fire between 6 to 45 Hz [7], or even at higher frequencies [9], depending on muscle type and contraction force or duration. These action potentials add to form the spectrum for the interference EMG signal of muscle contraction, of which most of the power is contained between 20 and 200 Hz [8], because of various physiological and technical factors increasing the range of frequencies [19,37]. Below 20 Hz, the content of the signal mainly corresponds to ‘‘movement artifact’’, i.e. movement at the electrode-skin interface caused by the muscle moving underneath the skin [8]. The tremor EMG envelope could therefore correspond to this ‘‘movement artifact’’, close to an ACC recording. However, the differential signal content of ACC and surface EMG envelope (obtained through a 12-Hz low-pass filter from raw signal recorded with a bandwidth of 20—450 Hz) has been demonstrated elsewhere [44]. The value of low-pass filtered EMG envelope to provide information regarding the electrical activity of the muscle involved in tremor, compared to the raw EMG signal, is further illustrated in Fig. 1. Finally, the signal-to-noise ratio (SNR) for ACC and EMG recordings in the different groups of subjects and patients was calculated as follows: SNR (in dB) = 10 × log10 (Pmax /mean [Poth ]), Pmax being the power at the peak frequency in the (0—15 Hz) range and mean (Poth ) being the mean power at the other frequencies.
Signal analysis ACC and EMG signals were analyzed first using FFT-based Welch’s algorithm [51], and then using EMD and HHT approaches. An FFT-based power spectral density (PSD) analysis was performed with the Matlab software on 10-second duration ACC recordings at the index finger and both EMG recordings at the forearm muscles using Welch’s algorithm with Hamming window of length 256 and 50% overlap between segments. Welch’s method shows significant reduction of variation compared with the periodograms, as it gives an averaged periodogram. In our work, the Matlab script ‘‘pwelch’’ was used with maximum precautions for the parameter settings to estimate the PSD [3]. The distinctive peaks corresponding to tremors in the (0—15 Hz) frequency range were well-conserved while eliminating the noiserelated spectrum and non-significant peaks. We determined whether the dominant frequency peak in the power spectrum provided by Welch’s algorithm was common or not
Please cite this article in press as: Ayache S-S, et al. Analysis of tremor in multiple sclerosis using Hilbert-Huang Transform. Neurophysiologie Clinique/Clinical Neurophysiology (2015), http://dx.doi.org/10.1016/j.neucli.2015.09.013
+Model NEUCLI-2502; No. of Pages 10
ARTICLE IN PRESS
4
S.-S. Ayache et al.
Figure 1 Envelope of the electromyographic (EMG) signal using a 0-15 Hz band-pass (top trace) corresponding to a raw EMG signal recorded using a 20-1,000 Hz band-pass (bottom trace). The low-pass filtered EMG envelope provides relevant information regarding the electrical activity of the muscle involved in tremor, compared to the raw EMG signal. Horizontal line: 100 ms, vertical line: 500 V.
between ACC and both EMG recordings. We concomitantly looked within a confidence interval of 0.3 Hz for common peaks in the PSDs of the three recordings. This confidence interval was chosen because a previous study showed that the frequency of most ‘‘organic’’ tremors could spontaneously fluctuate by about 0.2—0.3 Hz [38]. The representation of frequency changes as a function of time cannot be obtained precisely by FFT, even though by making use of the sliding windows and taking the FFT of each window, a change can be detected (which is in fact the Short Time Fourier Transform or spectrogram). Tremor may have an intermittent character, and therefore, not all windows may show tremor characteristics. In this context, time frequency analysis methods should be used, such as Hilbert Transform (HT) [34]. However, for non-stationary time series, a direct application of HT may produce erroneous and/or inconsistent results [49]. The EMD method, introduced by Huang et al. [26], is defined by an algorithm-based on empirical framework. It can be applied to non-linear and non-stationary biosignals. The EMD adaptively decomposes the signal, based on the time scale of the signal itself, into a finite number of monocomponent signals, called intrinsic mode functions (IMFs). The IMFs are symmetric and different IMFs yield different instantaneous local frequencies as a function of time, giving sharp identifications of embedded structures [26]. Signal decomposition to different modes by the EMD method leads to a more stable calculation of the Hilbert spectrum (HS) because IMFs admit well-behaved HT. The combination of EMD and HT is called Hilbert-Huang Transform (HHT) [24]. The HHT allows the HS to be obtained, in which both instantaneous amplitude and frequency of each IMF are functions of time in 3-D or 2-D plot. In 3-D plot, the amplitude (or the squared amplitude, also known as ‘‘energy’’) can be contoured over the instantaneous frequency-time plane as the z-axis surface. In 2-D plot, a color scale is used in the instantaneous frequency-time plane to represent amplitude or energy. The marginal Hilbert spectrum (MHS) is defined as the sum of the HSs obtained for the individual IMFs. The MHS offers a measure of total amplitude (or energy) contribution from each frequency value. It represents the cumulated amplitude (or energy) over the entire data span in a probabilistic sense and offers a measure of the total amplitude
(or energy) contribution from each frequency value, serving as an alternative spectrum expression of the data to the traditional Fourier spectrum [25]. Compared to the conventional FFT decomposition, this method gives a very sharp time resolution for the energy-frequency representation of the signal [15]. Thus, based on the above definitions, we performed three EMD procedures of signal analysis for the characterization and comparison of tremors. These procedures are described in detail in a previous paper [3] and they can be briefly summarized as follows. Procedure 1 was based on the application of the EMD algorithm and the determination of the IMF set for each ACC and EMG recording (Fig. 2). The power spectral densities (PSDs) of each IMF set were then estimated, based on Welch’s algorithm. For further analysis, we chose to keep the most significant IMFs in each recording, corresponding to the tremor-related activities of ACC and EMG recordings in the (3—8 Hz) frequency range (Fig. 3a and b). Our analysis was restricted to this frequency range, because MS tremor has been usually found within this range [2]. These activities were always contained within the first three IMFs [3]. We estimated concomitantly the PSDs (starting from the first PSD) corresponding to these IMFs. In the final step, we sought for common PSD peaks between ACC and both EMG recordings within a confidence interval of 0.3 Hz [3]. Procedure 2 was based on the estimation of HS of the significant IMFs in each ACC and EMG recording (Fig. 4). Then, we determined the normalized MHS of each HS and searched for common MHS peaks (Fig. 5). Procedure 3 was based on the determination of the maximum peak of the normalized MHSs obtained in each individual. Then, we calculated the average value of these maximum peaks in the four studied groups (MS patients with or without visible tremor, subjects with PT, and patients with ET). All individual and group average values were plotted in a graph against tremor frequency (Fig. 6).
Data analysis Descriptive statistics were expressed as mean ± standard deviation (SD). Statistical analyses were performed using InStat 3 (GraphPad Software, San Diego, CA, USA). Clinical
Please cite this article in press as: Ayache S-S, et al. Analysis of tremor in multiple sclerosis using Hilbert-Huang Transform. Neurophysiologie Clinique/Clinical Neurophysiology (2015), http://dx.doi.org/10.1016/j.neucli.2015.09.013
+Model NEUCLI-2502; No. of Pages 10
ARTICLE IN PRESS
EMD-HHT analysis of MS tremor
5
Figure 2 From top to bottom: example of the decomposition of a pre-filtered accelerometric (ACC) signal (dashed blue line) to its intrinsic mode functions (IMFs) (full red lines) in a patient with essential tremor.
data were compared between MS patients with or without visible tremor using the Mann-Whitney test for quantitative data and the Fisher’s exact test for categorical data. The influence of gender and age between the four studied groups (MS patients with or without visible tremor, subjects with PT, and patients with ET) was studied using a
Chi2 test and Kruskal-Wallis test with post-hoc Dunn’s multiple comparison tests, respectively. The SNRs of ACC and EMG recordings were compared using Friedman’s repeated measures Anova with post-hoc Dunn’s multiple comparison tests. Finally, tremor frequencies were compared between groups using one-way Anova with Tukey-Kramer multiple
Figure 3 From top to bottom: example of the decomposition of pre-filtered accelerometric (ACC) and electromyographic (EMG) signals to their most significant intrinsic mode functions (IMFs) (left panel, 3a) and the corresponding power spectral density of each IMF, based on Welch’s algorithm (right panel, 3b), in a patient with essential tremor. EMG1 and EMG2 refer to the EMG activity recorded from the extensor and flexor carpi radialis muscles, respectively.
Please cite this article in press as: Ayache S-S, et al. Analysis of tremor in multiple sclerosis using Hilbert-Huang Transform. Neurophysiologie Clinique/Clinical Neurophysiology (2015), http://dx.doi.org/10.1016/j.neucli.2015.09.013
+Model NEUCLI-2502; No. of Pages 10
ARTICLE IN PRESS
6
S.-S. Ayache et al.
Figure 4 From top to bottom: example of pre-filtered accelerometric (ACC) and electromyographic (EMG) signals (left panel) and the corresponding Hilbert spectra of their most significant intrinsic mode functions (right panel), in a patient with essential tremor. EMG1 and EMG2 refer to the EMG activity recorded from the extensor and flexor carpi radialis muscles, respectively.
Figure 5 From top to bottom: example of the normalized Marginal Hilbert Spectra of the accelerometric (ACC) and electromyographic (EMG) signals in a patient with essential tremor. EMG1 and EMG2 refer to the EMG activity recorded from the extensor and flexor carpi radialis muscles, respectively. A clear common peak can be identified between 7 and 8 Hz.
comparisons test. A P-value of less than 0.05 was considered significant in all cases.
Results Clinical comparisons The group of MS patients with visible tremor included 12 women and 4 men, aged from 31 to 74 years (mean ± SD:
49.6 ± 10.7) with EDSS ranging from 1.5 to 8.5 (4.4 ± 2.4). The group of MS patients without visible tremor included 6 women and 4 men, aged from 35 to 58 years (47.9 ± 6.6) with EDSS ranging from 2 to 6 (3.4 ± 1.6). The group of subjects with PT included 6 women and 11 men, aged from 23 to 63 years (35.3 ± 12.4). The group of patients with ET included 2 women and 6 men, aged from 62 to 87 years (71.4 ± 8.2). In all MS patients with visible tremor at clinical examination, this tremor involved both upper limbs, was present in attitude and during movement, but was totally absent at rest. Compared to MS patients without visible tremor, cerebellar signs were more frequent (94% vs. 50%, P = 0.018, Fisher’s test) and muscle strength at the upper limb was better (18.7 ± 1.6 vs 16.9 ± 2.1, P = 0.042, Mann-Whitney test) in MS patients with visible tremor. In contrast, functional impairment and disability, as assessed by EDSS, was similar in both groups (4.4 ± 2.4 vs 3.4 ± 1.6, P = 0.508). The four studied groups differed regarding gender (P = 0.049, Chi2 test), with 75% of MS with visible tremor being women and 75% of patients with ET being men. The groups also differed regarding age (P < 0.0001, Kruskal-Wallis test), patients with ET being older than the other patients and subjects (P < 0.05, Dunn’s test).
SNR The mean SNR (dB) ranged between 28.0 ± 2.8 (MS without tremor group) and 32.7 ± 5.2 (ET group) regarding ACC recordings, between 15.6 ± 4.1 (PT group) and 20.2 ± 6.9 (MS with tremor group) regarding EMG recordings in the extensor carpi radialis muscle, and between 19.7 ± 6.6 (ET group) and 21.2 ± 8.1 (MS without tremor group) regarding EMG recordings in the flexor carpi radialis muscle. Overall, the SNR was higher for ACC recording (29.9 dB ± 6.9) than for both EMG recordings (P < 0.0001, Friedman test and P < 0.001, Dunn’s tests). Conversely, the SNR did not significantly differ between the two EMG recordings (17.9 dB ± 5.8 vs 20.6 dB ± 7.1; P > 0.05, Dunn’s test).
Please cite this article in press as: Ayache S-S, et al. Analysis of tremor in multiple sclerosis using Hilbert-Huang Transform. Neurophysiologie Clinique/Clinical Neurophysiology (2015), http://dx.doi.org/10.1016/j.neucli.2015.09.013
+Model NEUCLI-2502; No. of Pages 10
ARTICLE IN PRESS
EMD-HHT analysis of MS tremor
7
Figure 6 Individual and mean (Me) values of the highest peaks in the three normalized Marginal Hilbert Spectra of each accelerometric and electromyographic recordings in the four studied groups: subjects with physiological tremor (PT), patients with essential tremor (ET), and multiple sclerosis patients without (MS NT) or with visible tremor (MS T).
FFT-based analysis Based on FFT analysis of ACC recordings at the index finger and both EMG recordings at the forearm muscles, a common dominant frequency peak was found in the (3—8 Hz) frequency range in all patients with ET, 2/17 subjects with PT, but no patients with MS, even in the presence of visible tremor (Table 1).
EMD- and MHS-based analysis Application of procedure 1 using EMD Procedure 1 revealed the presence of common frequency peaks in the (3—8 Hz) frequency range between ACC recordings at the index finger and both EMG recordings at the forearm muscles in 10/16 MS patients with visible tremor, 4/10 MS patients without visible tremor, 5/17 subjects with PT, and all patients with ET (Table 1). The prevalence of ACC-EMG coupling did not differ between MS patients with or without visible tremor (P = 0.42, Fisher’s test). However, the frequency differed between groups (MS patients with visible tremor: 4.6 Hz ± 0.9; MS patients without visible tremor: 3.8 Hz ± 0.5; subjects with PT: 3.8 Hz ± 0.7; patients with ET: 5.8 Hz ± 1.5; P = 0.004, one-way Anova). Post-hoc Tukey-Kramer multiple comparisons test showed that tremor frequency was higher in patients with ET compared to MS patients without visible tremor and subjects with PT (P < 0.05). Application of procedure 2 using MHS By applying procedure 2, a common frequency peak in the (3—8 Hz) frequency range between ACC recordings at the index finger and both EMG recordings at the forearm muscles was found in all patients with ET, but only one MS patient with visible tremor and one subject with PT (Table 1). Application of procedure 3 for group comparisons The maximum peaks of the normalized MHSs determined in each individual and the calculated average value of these maximum peaks in the four studied groups were plotted in a graph against tremor frequency (Fig. 6). This scattering plot showed a strong overlap between values observed in MS patients with or without visible tremor and subjects
with PT. The great majority of these values were of low frequency, mainly between 3 and 5 Hz. In contrast, there was a clear distinction between these values and those observed in patients with ET.
Discussion In this work, we performed ACC and EMG recordings at the upper limb in MS patients with or without visible action tremor. EMG recordings were used to highlight rhythmic activity in the envelope of the signal and not to analyze in detail the features and the recruitment of motor unit action potentials. The quality of the measured signal is often described by the SNR value. A SNR value as low as 6 dB is sufficient for the detection of tremor burst in EMG envelope [10], although it can be difficult to find coherence in a tremor when values are below 30 dB [22]. In our study, the mean SNRs ranged between about 28 and 33 dB for ACC and 16 and 21 dB for EMG. Despite these rather low SNRs compared to other studies [29], we have been able to find common ACC-EMG rhythmic activities, demonstrating the sensitivity of our methods of analysis. It cannot however be excluded that the prevalence of these common activities was underestimated due to these low SNRs. Finally, we found similar SNR values for EMG recordings performed in the extensor carpi radialis muscle and the flexor carpi radialis muscle, although the latter muscle is situated deep underneath the skin surface in the forearm position adopted during the recordings. Using a classical FFT-based PSD analysis, we did not find any common rhythmic activity between ACC recordings at the index finger and EMG recordings at forearm muscles in MS patients, even in the presence of a visible tremor. Conversely, a common ACC-EMG rhythmic activity was found in all patients with ET and two subjects with PT. The study of the normalized MHS (procedure 2) provided similar results, revealing a common frequency peak in all patients with ET, but only one subject with PT and also one MS patient with visible tremor. In contrast to the other methods, the combined EMD-HHT analysis (procedure 1) showed a common ACC-EMG frequency peak in 54% of MS patients (62.5% in case of visible tremor and 40% in the absence of visible tremor) and 29% of subjects with PT, in addition to all patients with ET. Tremor frequency was higher in the group of patients
Please cite this article in press as: Ayache S-S, et al. Analysis of tremor in multiple sclerosis using Hilbert-Huang Transform. Neurophysiologie Clinique/Clinical Neurophysiology (2015), http://dx.doi.org/10.1016/j.neucli.2015.09.013
+Model
ARTICLE IN PRESS
NEUCLI-2502; No. of Pages 10
8
S.-S. Ayache et al. Table 1 Common frequency peaks from accelerometric and electromyographic recordings using Fast Fourier Transform (FFT), Empirical Mode Decomposition (EMD), or marginal Hilbert spectrum (MHS) analysis. Participants Common peak in MS patients with visible tremor FFT EMD 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
4.05 4.3
MHS
Common peak in MS patients without visible tremor
Common peak in subjects with PT
FFT
FFT
EMD
MHS
EMD
Common peak in patients with ET MHS
4.5 10
6.9 4.65 4 4.2
3.7 3.55 4.45
3.3
3.45 7.9
3.75 5.4 4.15 3.75 and 4.95
3.4 4.9
FFT
EMD
MHS
6.2 5.9 5.6 6.3 8.1 5.9 6.1 7.3
6.3 4.00 and 7.05 5.55 3.85 and 7.00 7.45 5.95 6.1 3.15 and 7.00
6 6 5.8 6 8 6 6 7
8
3.35 3.85
MS: multiple sclerosis; PT: physiological tremor; ET: essential tremor; ‘‘ ’’ means that no common peak was found; the values have been rounded to 0.05 or 0.10 Hz.
with ET than in the other groups, as revealed by the values of the maximum peaks of the normalized MHSs. The usual definition of tremor is purely clinical, i.e. a ‘‘rhythmical, involuntary oscillatory movement of a body part’’ [13]. Therefore, a single-channel ACC recording is sufficient to show such a ‘‘rhythmic, oscillatory activity’’ defining tremor and this approach is commonly used for the diagnosis of various types of tremor [6,23,27,47,52], including MS tremor [2]. However, to demonstrate that a pathological tremor arises from a central generator, it is crucial to show EMG synchronization at the accelerometric tremor frequency, meaning that ACC-EMG coupling drives the rhythmic movements of the tremulous limb [12,14,16,42,43]. Actually, FFT and MHS analyses failed to detect ACC-EMG coupling in almost all MS patients in our series. Conversely, EMD-HHT analysis revealed ACC-EMG coupling in 54% of MS patients, supporting the value of this procedure of signal analysis in combination with multichannel ACC-EMG recordings to reveal the presence of centrally driven pathological tremor. The second objective of this study was to compare the characteristics of MS tremor to those of ET and PT. These characteristics were very similar in MS patients (with or without visible tremor) and in subjects with PT. In both groups, we found a rather inconstant tremor at a quite low frequency, whereas patients with ET had a highly constant tremor at higher frequency. In most previously published studies, PT was studied by means of single-channel ACC recording, favoring high frequency components of tremor. Carignan et al. [5] were the first to reveal the importance of low frequency components in PT and their close
correlation with tremor amplitude. Using EMD-HHT analysis of ACC-EMG coupling, we confirmed the major impact of low-frequency components in PT. Similar low-frequency components were found in MS patients but of higher amplitude. Thus, in most MS patients, tremor could correspond to exaggerated PT, which has not been previously considered as a leading hypothesis, except in one recent study [35]. In this latter study, the authors showed that the principal ACC features of postural tremor recorded at index finger were similar in healthy individuals and MS patients, although the oscillatory output was of greater amplitude, more regular and had a higher degree of coupling in MS patients. They concluded: ‘‘this similarity suggests that, despite the neurological damage caused by this disease, the oscillators underlying tremorgenesis for the MS and healthy adults were essentially the same’’. Our results are in agreement with this conclusion. One intriguing observation was that ACC-EMG characteristics and the prevalence of ACC-EMG coupling were similar in the two groups of MS patients whether tremor was ‘‘visible’’ or not. In fact, MS patients with ‘‘visible tremor’’ only differed from MS patients without ‘‘visible tremor’’ by higher prevalence of cerebellar signs and greater muscle strength at the upper limb. Thus, MS tremor could be a form of PT whose ‘‘visibility’’ is enhanced by cerebellar dysfunction in the absence of motor deficit. Conversely, the presence of motor weakness could reduce the visibility of PT. In fact, PT is a ubiquitous property of the neuromuscular system that becomes exaggerated and more visible in certain conditions like fatigue, stress, and hypothyroidism, among others. Cerebellar dysfunction in MS patients could
Please cite this article in press as: Ayache S-S, et al. Analysis of tremor in multiple sclerosis using Hilbert-Huang Transform. Neurophysiologie Clinique/Clinical Neurophysiology (2015), http://dx.doi.org/10.1016/j.neucli.2015.09.013
+Model NEUCLI-2502; No. of Pages 10
ARTICLE IN PRESS
EMD-HHT analysis of MS tremor be one of these conditions, since cerebellum plays a key role in the control of movement. In addition, in the case of cerebellar dysfunction, moving a limb to an intended target or maintaining it in a given position leads to continuous adjustments and readjustments, which can be mistaken for tremor. It is likely that cases of ‘‘visible tremor’’ noticed in MS patients could in fact reveal the impact of dysmetria or cerebellar ataxia. In a previous report, Sabra and Hallett [46] expressed the same opinion that ‘‘dissecting intention tremor from serial dysmetria and postural tremor from ataxia is difficult’’. The term ‘‘pseudotremor’’ covers pathological conditions such as dysmetria and ataxia that produce pseudorhythmic movements mimicking tremor in patients with neurological disorders. The high prevalence of cerebellar alterations in MS patients may be a primary cause of tremor-like activity in these patients [1,45,50].
Conclusion To summarize, our results suggest that ‘‘visible tremor’’ in MS patients could correspond to an exaggerated PT (with ACC-EMG coupling) or a pseudorhythmic activity (‘‘pseudotremor’’) related to dysmetria or ataxia (without ACC-EMG coupling). Both conditions are favored by cerebellar dysfunction, but cannot be considered strictly speaking as forms of ‘‘pathological tremor’’, directly generated by a central oscillator that would be originally created by MS lesions. We cannot rule out however the possible contribution of a low-frequency tremor originating from lesions in the brainstem (midbrain) or cerebellothalamic circuits, but such a ‘‘pathological tremor’’ may be much less frequent than expected in MS. The limitations of this study were a small sample size of heterogeneous MS patients and the absence of neuroimaging assessment. Therefore, further studies should be designed to confirm these results in a larger population of MS patients and to correlate ACC-EMG features to the location of lesions or metabolic changes into the brain. On clinical examination, it is difficult to characterize MS tremor perfectly. Multichannel ACC-EMG recordings with EMD-HHT analysis may help to differentiate which is related to an exaggerated PT, a ‘‘pseudotremor’’, or a ‘‘pathological tremor’’. This better characterization might be important to distinguish different underlying pathophysiological mechanisms that can respond differentially to treatment, as previously suggested by the predictive value of tremor analysis in MS patients treated by thalamotomy [32]. Finally, the real value of our EMD-HHT approach is also to be able to reveal surface EMG-ACC coupling when the SNR of these recordings is low.
Disclosure of interest The authors declare that they have no competing interest.
References [1] Alusi SH, Glickman S, Aziz TZ, Bain PG. Tremor in multiple sclerosis. J Neurol Neurosurg Psychiatry 1999;66:131—4. [2] Alusi SH, Worthington J, Glickman S, Bain PG. A study of tremor in multiple sclerosis. Brain 2001;124:720—30.
9 [3] Ayache SS, Al-Ani T, Lefaucheur JP. Distinction between essential and physiological tremor using Hilbert-Huang Transform (HHT). Neurophysiol Clin 2014;44:203—12. [4] Barreca SR, Stratford PW, Lambert CL, Masters LM, Streiner DL. Test-retest reliability, validity, and sensitivity of the Chedoke arm and hand activity inventory: a new measure of upperlimb function for survivors of stroke. Arch Phys Med Rehabil 2005;86:1616—22. [5] Carignan B, Daneault JF, Duval C. Quantifying the importance of high frequency components on the amplitude of physiological tremor. Exp Brain Res 2010;202:299—306. [6] Costa J, González HA, Valldeoriola F, Gaig C, Tolosa E, Valls-Solé J. Nonlinear dynamic analysis of oscillatory repetitive movements in Parkinson’s disease and essential tremor. Mov Disord 2010;25:2577—86. [7] De Luca CJ, Contessa P. Hierarchical control of motor units in voluntary contractions. J Neurophysiol 2012;107:178—95. [8] De Luca CJ, Gilmore LD, Kuznetsov M, Roy SH. Filtering the surface EMG signal: movement artifact and baseline noise contamination. J Biomech 2010;43:1573—9. [9] De Luca CJ, Hostage EC. Relationship between firing rate and recruitment threshold of motoneurons in voluntary isometric contractions. J Neurophysiol 2010;104:1034—46. [10] De Marchis C, Conforto S, Severini G, Schmid M, D’Alessio T. Detection of tremor bursts from the sEMG signal: an optimization procedure for different detection methods. Conf Proc IEEE Eng Med Biol Soc 2011;2011:7508—11. [11] Deuschl G, Lauk M, Timmer J. Tremor classification and tremor time series analysis. Chaos 1995;5:48—51. [12] Deuschl G, Krack P, Lauk M, Timmer J. Clinical neurophysiology of tremor. J Clin Neurophysiol 1996;13:110—21. [13] Deuschl G, Bain P, Brin M, Ad Hoc Scientific Committee. Consensus statement of the Movement Disorder Society on tremor. Mov Disord 1998;13(Suppl. 3):2—23. [14] Deuschl G, Raethjen J, Lindemann M, Krack P. The pathophysiology of tremor. Muscle Nerve 2001;24:716—35. [15] Donnelly D. The Fast Fourier and Hilbert-Huang Transforms: a comparison. IJCCC 2006;1:45—52. [16] Elble RJ. Characteristics of physiologic tremor in young and elderly adults. Clin Neurophysiol 2003;114:624—35. [17] Elble RJ, Randall JE. Motor-unit activity responsible for 8- to 12-Hz component of human physiological finger tremor. J Neurophysiol 1976;39:370—83. [18] Elble RJ, Randall JE. Mechanistic components of normal hand tremor. Electroencephalogr Clin Neurophysiol 1978;44: 72—82. [19] Farina D, Merletti R, Enoka RM. The extraction of neural strategies from the surface EMG: an update. J Appl Physiol (1985) 2014;117:1215—30. [20] Fox JR, Randall JE. Relationship between forearm tremor and the biceps electromyogram. J Appl Physiol 1970;29: 103—8. [21] Gironell A, Kulisevsky J, Pascual-Sedano B, Barbanoj M. Routine neurophysiologic tremor analysis as a diagnostic tool for essential tremor: a prospective study. J Clin Neurophysiol 2004;21:446—50. [22] Hellwig B, Häussler S, Schelter B, Lauk M, Guschlbauer B, Timmer J, et al. Tremor-correlated cortical activity in essential tremor. Lancet 2001;357:519—23. [23] Hossen A, Muthuraman M, Al-Hakim Z, Raethjen J, Deuschl G, Heute U. Discrimination of Parkinsonian tremor from essential tremor using statistical signal characterization of the spectrum of accelerometer signal. Biomed Mater Eng 2013;23: 513—31. [24] Huang NE. Introduction to Hilbert-Huang Transform and its related mathematical problems. In: Huang NE, Shen SSP, editors. Hilbert-Huang Transform and its applications. Singapore: World Scientific; 2005. p. 1—26.
Please cite this article in press as: Ayache S-S, et al. Analysis of tremor in multiple sclerosis using Hilbert-Huang Transform. Neurophysiologie Clinique/Clinical Neurophysiology (2015), http://dx.doi.org/10.1016/j.neucli.2015.09.013
+Model NEUCLI-2502; No. of Pages 10
ARTICLE IN PRESS
10 [25] Huang NE, Wu Z. A review on Hilbert-Huang Transform: method and its applications to geophysical studies. Rev Geophys 2008;46:RG2006. [26] Huang NE, Shen Z, Long SR, Wu MC, Shih HH, Zheng Q, et al. The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis. Proc R Soc London 1998;454:903—95. [27] Jang W, Han J, Park J, Kim JS, Cho JW, Koh SB, et al. Waveform analysis of tremor may help to differentiate Parkinson’s disease from drug-induced parkinsonism. Physiol Meas 2013;34:N15—24. [28] Koch M, Mostert J, Heersema D, De Keyser J. Tremor in multiple sclerosis. J Neurol 2007;254:133—45. [29] Kugler P, Jaremenko C, Schlachetzki J, Winkler J, Klucken J, Eskofier B. Automatic recognition of Parkinson’s disease using surface electromyography during standardized gait tests. Conf Proc IEEE Eng Med Biol Soc 2013;2013:5781—4. [30] Kurtzke JF. Rating neurologic impairment in multiple sclerosis: an expanded disability status scale (EDSS). Neurology 1983;33:1444—52. [31] Li K, Hogrel JY, Duchêne J, Hewson DJ. Analysis of fatigue and tremor during sustained maximal grip contractions using Hilbert-Huang Transformation. Med Eng Phys 2012;34: 832—40. [32] Liu X, Aziz TZ, Miall RC, Rowe J, Alusi SH, Bain PG, et al. Frequency analysis of involuntary movements during wrist tracking: a way to identify ms patients with tremor who benefit from thalamotomy. Stereotact Funct Neurosurg 2000;74:53—62. [33] Medical Research Council. Aids to examination of the peripheral nervous system. In: Memorandum no. 45. London: Her Majesty’s Stationary Office; 1976. [34] Mitra PP, Pesaran B. Analysis of dynamic brain imaging data. Biophys J 1999;76:691—708. [35] Morrison S, Sosnoff JJ, Sandroff BM, Pula JH, Motl RW. The dynamics of finger tremor in multiple sclerosis is affected by whole body position. J Neurol Sci 2013;324:84—9. [36] Muthuraman M, Hossen A, Heute U, Deuschl G, Raethjen J. A new diagnostic test to distinguish tremulous Parkinson’s disease from advanced essential tremor. Mov Disord 2011;26: 1548—52. [37] Myers LJ, Lowery M, O’Malley M, Vaughan CL, Heneghan C, St Clair Gibson A, et al. Rectification and non-linear preprocessing of EMG signals for corticomuscular analysis. J Neurosci Methods 2003;124:157—65. [38] O’Suilleabhain PE, Matsumoto JY. Time-frequency analysis of tremors. Brain 1998;121:2127—34.
S.-S. Ayache et al. [39] Palit SK, Mukherjee S, Bhattacharya DK. New types of nonlinear autocorrelations of bivariate data and their applications. Appl Math Comput 2012;218:8951—67. [40] Pittock SJ, McClelland RL, Mayr WT, Rodriguez M, Matsumoto JY. Prevalence of tremor in multiple sclerosis and associated disability in the Olmsted County population. Mov Disord 2004;19:1482—5. [41] Polman CH, Reingold SC, Banwell B, Clanet M, Cohen JA, Filippi M, et al. Diagnostic criteria for multiple sclerosis: 2010 revisions to the McDonald criteria. Ann Neurol 2011;69: 292—302. [42] Raethjen J, Pawlas F, Lindemann M, Wenzelburger R, Deuschl G. Determinants of physiologic tremor in a large normal population. Clin Neurophysiol 2000;111:1825—37. [43] Raethjen J, Lauk M, Köster B, Fietzek U, Friege L, Timmer J, et al. Tremor analysis in two normal cohorts. Clin Neurophysiol 2004;115:2151—6. [44] Roy SH, Cheng MS, Chang SS, Moore J, De Luca G, Nawab SH, et al. A combined sEMG and accelerometer system for monitoring functional activity in stroke. IEEE Trans Neural Syst Rehabil Eng 2009;17:585—94. [45] Ruutiainen J. Assessment and treatment of ataxia in multiple sclerosis. In: Ketelaer P, Prosiegel M, Battaglia M, Messmer U, editors. A problem-oriented approach to multiple sclerosis. Leuven: Amersfoort; 1997. p. 227—35. [46] Sabra AF, Hallett M. Action tremor with alternating activity in antagonist muscles. Neurology 1984;34:151—6. [47] Shaikh AG, Jinnah HA, Tripp RM, Optican LM, Ramat S, Lenz FA, et al. Irregularity distinguishes limb tremor in cervical dystonia from essential tremor. J Neurol Neurosurg Psychiatry 2008;79:187—9. [48] Timmer J, Haussler S, Lauk M, Lucking CH. Pathological tremors: deterministic chaos or nonlinear stochastic oscillators? Chaos 2000;10:278—88. [49] Veltcheva A, Guedes SC. Identification of the components of wave spectra by the Hilbert Huang Transform method. Appl Ocean Res 2004;26:1—12. [50] Weinshenker BG, Issa M, Baskerville J. Long-term and shortterm outcome of multiple sclerosis: a 3-year follow-up study. Arch Neurol 1996;53:353—8. [51] Welch P. The use of Fast Fourier Transform for the estimation of power spectra: a method based on time averaging over short, modified periodograms. IEEE Trans Audio Electroacoustics 1967;15:70—3. [52] Zeuner KE, Shoge RO, Goldstein SR, Dambrosia JM, Hallett M. Accelerometry to distinguish psychogenic from essential or parkinsonian tremor. Neurology 2003;61:548—50.
Please cite this article in press as: Ayache S-S, et al. Analysis of tremor in multiple sclerosis using Hilbert-Huang Transform. Neurophysiologie Clinique/Clinical Neurophysiology (2015), http://dx.doi.org/10.1016/j.neucli.2015.09.013
ARTICLE IN PRESS
Neurophysiologie Clinique/Clinical Neurophysiology (2015) xxx, xxx—xxx
Disponible en ligne sur
ScienceDirect www.sciencedirect.com
ORIGINAL ARTICLE/ARTICLE ORIGINAL
Analysis of tremor in multiple sclerosis using Hilbert-Huang Transform Analyse du tremblement de la sclérose en plaques utilisant la transformée de Hilbert-Huang S.-S. Ayache a,b, T. Al-ani a, W.-H. Farhat a,b, H.-G. Zouari a,b,c, A. Créange a,d, J.-P. Lefaucheur a,b,∗ a
EA4391, faculté de médecine de Créteil, université Paris Est—Créteil, 8, avenue du Général-Sarrail, 94010 Créteil cedex, France b Service de physiologie et d’explorations fonctionnelles, hôpital Henri-Mondor, Assistance publique—Hôpitaux de Paris, 51, avenue du Maréchal-de-Lattre-de-Tassigny, 94010 Créteil cedex, France c Service d’explorations fonctionnelles, CHU Habib Bourguiba, Sfax, Tunisia d Service de neurologie, hôpital Henri-Mondor, Assistance publique—Hôpitaux de Paris, 51, avenue du Maréchal-de-Lattre-de-Tassigny, 94010 Créteil cedex, France Received 15 February 2015; accepted 27 September 2015
KEYWORDS Accelerometer; Action tremor; Electromyography; Empirical mode decomposition; Multiple sclerosis; Signal analysis
∗
Summary Tremor is frequently described in patients with multiple sclerosis (MS) but remains poorly characterized using neurophysiological techniques. Accelerometric (ACC) and electromyographic (EMG) recordings were performed in 26 MS patients complaining of clumsiness, associated (n = 16) or not associated (n = 10) with visible tremor. Seventeen healthy subjects with physiological tremor (PT) and eight patients with essential tremor (ET) served as controls. Signals were analyzed using non-linear Empirical Mode Decomposition (EMD) and related HilbertHuang Transform (HHT), compared to the standard linear spectral analysis using Fast Fourier Transform (FFT). The presence of cerebellar signs and motor deficit was assessed on clinical examination. Using FFT, tremor was found in all patients with ET and 12% of subjects with PT, but in none of the MS patients, even in the presence of visible tremor. In contrast, EMD-HHT analysis of ACC-EMG coupling showed common frequency peaks characterizing tremor related to a central generator in 62.5% of MS patients with visible tremor, 40% of MS patients without visible tremor, 29% of subjects with PT, and all patients with ET. In EMD-HHT analysis, tremor characteristics were similar in subjects with PT and MS patients, regardless of the presence of a visible tremor, but these characteristics clearly differed in patients with ET. A visible tremor in MS patients was associated with more frequent cerebellar signs and less motor deficit at
Corresponding author. E-mail address: [email protected] (J.-P. Lefaucheur).
http://dx.doi.org/10.1016/j.neucli.2015.09.013 0987-7053/© 2015 Elsevier Masson SAS. All rights reserved.
Please cite this article in press as: Ayache S-S, et al. Analysis of tremor in multiple sclerosis using Hilbert-Huang Transform. Neurophysiologie Clinique/Clinical Neurophysiology (2015), http://dx.doi.org/10.1016/j.neucli.2015.09.013
+Model NEUCLI-2502; No. of Pages 10
ARTICLE IN PRESS
2
S.-S. Ayache et al. the upper limb. The low-frequency tremor observed in MS patients could therefore originate in lesions of the brainstem (midbrain) or cerebellothalamic circuits, or may correspond to an enhanced PT, partly favored by cerebellar dysfunction and being more visible during movement execution in the absence of concomitant motor deficit. © 2015 Elsevier Masson SAS. All rights reserved.
MOTS CLÉS Accéléromètre ; Analyse du signal ; Décomposition en mode empirique ; Électromyographie ; Sclérose en plaques ; Tremblement d’action
Résumé Le tremblement est souvent décrit chez les patients atteints de sclérose en plaques (SEP), mais reste mal caractérisé sur le plan neurophysiologique. Des enregistrements accélérométriques (ACC) et électromyographiques (EMG) ont été réalisés chez 26 patients atteints de SEP qui se plaignaient de maladresse, associée (n = 16) ou non (n = 10) à un tremblement visible cliniquement. Dix-sept sujets sains présentant un tremblement physiologique (TP) et huit patients atteints de tremblement essentiel (TE) ont servi de témoins. Les signaux ont été analysés par une méthode non-linéaire de décomposition en mode empirique (EMD) associée à une transformée de Hilbert-Huang (HHT), comparée à l’analyse spectrale linéaire classique, utilisant une transformée rapide de Fourier (FFT). La présence de signes cérébelleux et d’un déficit moteur a été évaluée cliniquement. En utilisant la FFT, un tremblement a été trouvé chez tous les patients avec TE et 12 % des sujets avec TP, mais chez aucun patient atteint de SEP, même en présence d’un tremblement visible cliniquement. En revanche, l’analyse EMD-HHT du couplage ACC-EMG a montré des pics de fréquence communs témoignant d’un tremblement avec générateur central chez 62,5 % des patients atteints de SEP avec tremblement visible, 40 % des patients atteints de SEP sans tremblement visible, 29 % des sujets avec TP, et chez tous les patients avec TE. Dans cette analyse EMD-HHT, les caractéristiques du tremblement étaient similaires entre les sujets avec TP et les patients atteints de SEP, indépendamment de la présence d’un tremblement visible, mais il est clair que ces caractéristiques différaient chez les patients avec TE. Le tremblement visible chez les patients atteints de SEP était associé à des signes cérébelleux plus fréquents et à un moindre déficit moteur. Ainsi, le tremblement de basse fréquence observé chez les patients atteints de SEP pourrait avoir pour origine une lésion du tronc cérébral (mésencéphale) ou de circuits cérébellothalamiques ou bien correspondre à un TP exagéré, en partie favorisé par une dysfonction cérébelleuse et que l’absence de déficit moteur concomitant rendrait plus visible cliniquement pendant l’exécution d’un mouvement. © 2015 Elsevier Masson SAS. Tous droits réservés.
Introduction Tremor was found to be clinically present in 25% and 58% of patients with multiple sclerosis (MS) in two previously published large cohorts [2,40]. In MS patients, tremor usually consists of an action tremor of the upper extremities, the rest component being almost inexistent [2]. However, MS patients may have various types of action tremor intermingled with other movement disorders in a complex clinical picture [1,2,28]. It may be difficult to characterize tremor in this context and neurophysiological investigations can be helpful. Clinical neurophysiology of tremor is usually based on the analysis of accelerometric (ACC) and electromyographic (EMG) recordings with Fast Fourier Transform (FFT), including the study of peaks in the power spectral analysis of periodograms, regularity in the time-frequency domain, asymmetry of the autocorrelation function decay, or coupling by applying cross-correlation and coherence analyses [11,21,36,38]. Despite the contribution of FFT methods in this domain, shown in these and other works, the appropriateness of using these methods is still limited since it is based on the assumption that tremor dynamics reflect periodic and linear activity, which is in fact not true [39,48]. Indeed tremor is a non-periodic, non-linear and non-stationary
signal with time-varying frequency, leaving room for nonlinear methods of signal analysis. Another limitation of FFT application is that it cannot distinguish between different types of tremor, whose frequency overlaps in similar ranges. In this context, Empirical Mode Decomposition (EMD) [26] and related Hilbert-Huang Transform (HHT) [24] could be relevant, according to their ability to deal with a nonstationary signal and their capacity to detect low-frequency components embedded in a non-stationary background [31]. In a previous pilot study including healthy subjects and patients with essential tremor (ET) [3], we described various EMD-HHT procedures for tremor analysis. The objective of the present study was to determine whether these procedures could be applied in MS patients to better characterize tremor in this particular clinical condition. In addition, tremor characteristics observed in MS patients were compared to those of other types of tremor (physiological tremor [PT] and ET), using the same signal analysis approach.
Methods Participants Sixteen MS patients with visible tremor in one or both upper extremities and ten MS patients without visible tremor were
Please cite this article in press as: Ayache S-S, et al. Analysis of tremor in multiple sclerosis using Hilbert-Huang Transform. Neurophysiologie Clinique/Clinical Neurophysiology (2015), http://dx.doi.org/10.1016/j.neucli.2015.09.013
+Model NEUCLI-2502; No. of Pages 10
ARTICLE IN PRESS
EMD-HHT analysis of MS tremor enrolled from the Department of Neurology of Henri Mondor hospital. All these patients complained of clumsiness in one or both upper limbs. Clumsiness was defined as the incapacity to perform at least one of the daily life activities listed in the Chedoke Arm and Hand Activity Inventory [4]. Patients were also selected according to the following criteria: • confirmed diagnosis of MS in reference to the revised McDonald’s criteria [41]; • no history of MS relapse in the previous month; • age between 18 and 70 years; • absence of any other neurological or psychiatric disease; • signed informed consent. Two groups of patients served as controls: a first group of 17 healthy subjects with PT and a second group of eight patients with ET. PT and ET were defined according to the consensus statement of Movement Disorder Society on tremor [13].
Clinical neurological examination MS patients underwent standard neurological examination with special attention paid to the most affected upper limb, to determine the presence or absence of a visible tremor, cerebellar signs (dysdiadochokinesis and dysmetria), and motor deficit. For the latter purpose, we calculated a motor score ranging from 0 (null strength) to 20 (full strength), as the summation of the scores measured for four muscle groups (pinch, wrist extension, forearm flexion and arm abduction) using the Medical Research Council (MRC) scale for muscle strength [33]. In addition, the Expanded Disability Status Scale (EDSS) was scored for each MS patient [30].
Neurophysiological investigation The same investigator (SSA) performed neurophysiological investigation, consisting of ACC and EMG recordings. Patients and subjects were seated in a comfortable chair with the upper limb fully at rest or held in the ‘‘oath position’’. Tremor evaluation was carried out on the side most severely affected by tremor or clumsiness. If both sides were equally affected, we chose to evaluate the right one. For ACC recordings, a piezoelectric single plane accelerometer (TREM0000, Neuroservices, Evry-Lisses, France) was fixed on the dorsal aspect of the distal phalanx of the index finger. Bipolar EMG was recorded from the extensor and flexor carpi radialis muscles, using a pair of surface pre-gelled electrodes (Ref 9013S0242, NatusDantec, Skovlunde, Danemark), placed 2 cm apart over the muscle belly (see the picture of the experimental setup in [3]). Thus, both ACC and EMG recordings concerned the same movement (extension/flexion of the hand and fingers). All recordings were performed with a Keypoint machine (Natus-Dantec). Signals were amplified, filtered (band-pass: 1—20 Hz for ACC signals; 20—1000 Hz for EMG signals) and uniformly sampled at 24 kHz. To shorten the computation time, the ACC and EMG data were downsampled to 1000 Hz using a polyphase implementation in Matlab software (MathWorks, Natick, MA, USA). The resampled signals were then filtered over the (0—15 Hz) frequency range. This offline
3 analysis allowed study of the envelope of the tremor signal, according to the method developed by Fox and Randall [20], then used, for example, by Elble and Randall [17,18] or O’Suilleabhain and Matsumoto [38]. Prior to frequency analysis, these authors low-pass filtered at 20—21 Hz the raw EMG signal recorded with a band-pass of 10—3000 Hz or 30—500 Hz. The resulting envelope of the surface EMG provides a measure of any rhythmical variations within the frequency band of tremor [17,18]. Similarly, in the present study, we looked at the tremor envelope of the EMG signal and not at the pattern of motor unit recruitment during voluntary muscle contraction, whose frequency analysis requires higher and broader band-pass. Actually, during voluntary muscle contraction, it is known that single motor unit action potentials fire between 6 to 45 Hz [7], or even at higher frequencies [9], depending on muscle type and contraction force or duration. These action potentials add to form the spectrum for the interference EMG signal of muscle contraction, of which most of the power is contained between 20 and 200 Hz [8], because of various physiological and technical factors increasing the range of frequencies [19,37]. Below 20 Hz, the content of the signal mainly corresponds to ‘‘movement artifact’’, i.e. movement at the electrode-skin interface caused by the muscle moving underneath the skin [8]. The tremor EMG envelope could therefore correspond to this ‘‘movement artifact’’, close to an ACC recording. However, the differential signal content of ACC and surface EMG envelope (obtained through a 12-Hz low-pass filter from raw signal recorded with a bandwidth of 20—450 Hz) has been demonstrated elsewhere [44]. The value of low-pass filtered EMG envelope to provide information regarding the electrical activity of the muscle involved in tremor, compared to the raw EMG signal, is further illustrated in Fig. 1. Finally, the signal-to-noise ratio (SNR) for ACC and EMG recordings in the different groups of subjects and patients was calculated as follows: SNR (in dB) = 10 × log10 (Pmax /mean [Poth ]), Pmax being the power at the peak frequency in the (0—15 Hz) range and mean (Poth ) being the mean power at the other frequencies.
Signal analysis ACC and EMG signals were analyzed first using FFT-based Welch’s algorithm [51], and then using EMD and HHT approaches. An FFT-based power spectral density (PSD) analysis was performed with the Matlab software on 10-second duration ACC recordings at the index finger and both EMG recordings at the forearm muscles using Welch’s algorithm with Hamming window of length 256 and 50% overlap between segments. Welch’s method shows significant reduction of variation compared with the periodograms, as it gives an averaged periodogram. In our work, the Matlab script ‘‘pwelch’’ was used with maximum precautions for the parameter settings to estimate the PSD [3]. The distinctive peaks corresponding to tremors in the (0—15 Hz) frequency range were well-conserved while eliminating the noiserelated spectrum and non-significant peaks. We determined whether the dominant frequency peak in the power spectrum provided by Welch’s algorithm was common or not
Please cite this article in press as: Ayache S-S, et al. Analysis of tremor in multiple sclerosis using Hilbert-Huang Transform. Neurophysiologie Clinique/Clinical Neurophysiology (2015), http://dx.doi.org/10.1016/j.neucli.2015.09.013
+Model NEUCLI-2502; No. of Pages 10
ARTICLE IN PRESS
4
S.-S. Ayache et al.
Figure 1 Envelope of the electromyographic (EMG) signal using a 0-15 Hz band-pass (top trace) corresponding to a raw EMG signal recorded using a 20-1,000 Hz band-pass (bottom trace). The low-pass filtered EMG envelope provides relevant information regarding the electrical activity of the muscle involved in tremor, compared to the raw EMG signal. Horizontal line: 100 ms, vertical line: 500 V.
between ACC and both EMG recordings. We concomitantly looked within a confidence interval of 0.3 Hz for common peaks in the PSDs of the three recordings. This confidence interval was chosen because a previous study showed that the frequency of most ‘‘organic’’ tremors could spontaneously fluctuate by about 0.2—0.3 Hz [38]. The representation of frequency changes as a function of time cannot be obtained precisely by FFT, even though by making use of the sliding windows and taking the FFT of each window, a change can be detected (which is in fact the Short Time Fourier Transform or spectrogram). Tremor may have an intermittent character, and therefore, not all windows may show tremor characteristics. In this context, time frequency analysis methods should be used, such as Hilbert Transform (HT) [34]. However, for non-stationary time series, a direct application of HT may produce erroneous and/or inconsistent results [49]. The EMD method, introduced by Huang et al. [26], is defined by an algorithm-based on empirical framework. It can be applied to non-linear and non-stationary biosignals. The EMD adaptively decomposes the signal, based on the time scale of the signal itself, into a finite number of monocomponent signals, called intrinsic mode functions (IMFs). The IMFs are symmetric and different IMFs yield different instantaneous local frequencies as a function of time, giving sharp identifications of embedded structures [26]. Signal decomposition to different modes by the EMD method leads to a more stable calculation of the Hilbert spectrum (HS) because IMFs admit well-behaved HT. The combination of EMD and HT is called Hilbert-Huang Transform (HHT) [24]. The HHT allows the HS to be obtained, in which both instantaneous amplitude and frequency of each IMF are functions of time in 3-D or 2-D plot. In 3-D plot, the amplitude (or the squared amplitude, also known as ‘‘energy’’) can be contoured over the instantaneous frequency-time plane as the z-axis surface. In 2-D plot, a color scale is used in the instantaneous frequency-time plane to represent amplitude or energy. The marginal Hilbert spectrum (MHS) is defined as the sum of the HSs obtained for the individual IMFs. The MHS offers a measure of total amplitude (or energy) contribution from each frequency value. It represents the cumulated amplitude (or energy) over the entire data span in a probabilistic sense and offers a measure of the total amplitude
(or energy) contribution from each frequency value, serving as an alternative spectrum expression of the data to the traditional Fourier spectrum [25]. Compared to the conventional FFT decomposition, this method gives a very sharp time resolution for the energy-frequency representation of the signal [15]. Thus, based on the above definitions, we performed three EMD procedures of signal analysis for the characterization and comparison of tremors. These procedures are described in detail in a previous paper [3] and they can be briefly summarized as follows. Procedure 1 was based on the application of the EMD algorithm and the determination of the IMF set for each ACC and EMG recording (Fig. 2). The power spectral densities (PSDs) of each IMF set were then estimated, based on Welch’s algorithm. For further analysis, we chose to keep the most significant IMFs in each recording, corresponding to the tremor-related activities of ACC and EMG recordings in the (3—8 Hz) frequency range (Fig. 3a and b). Our analysis was restricted to this frequency range, because MS tremor has been usually found within this range [2]. These activities were always contained within the first three IMFs [3]. We estimated concomitantly the PSDs (starting from the first PSD) corresponding to these IMFs. In the final step, we sought for common PSD peaks between ACC and both EMG recordings within a confidence interval of 0.3 Hz [3]. Procedure 2 was based on the estimation of HS of the significant IMFs in each ACC and EMG recording (Fig. 4). Then, we determined the normalized MHS of each HS and searched for common MHS peaks (Fig. 5). Procedure 3 was based on the determination of the maximum peak of the normalized MHSs obtained in each individual. Then, we calculated the average value of these maximum peaks in the four studied groups (MS patients with or without visible tremor, subjects with PT, and patients with ET). All individual and group average values were plotted in a graph against tremor frequency (Fig. 6).
Data analysis Descriptive statistics were expressed as mean ± standard deviation (SD). Statistical analyses were performed using InStat 3 (GraphPad Software, San Diego, CA, USA). Clinical
Please cite this article in press as: Ayache S-S, et al. Analysis of tremor in multiple sclerosis using Hilbert-Huang Transform. Neurophysiologie Clinique/Clinical Neurophysiology (2015), http://dx.doi.org/10.1016/j.neucli.2015.09.013
+Model NEUCLI-2502; No. of Pages 10
ARTICLE IN PRESS
EMD-HHT analysis of MS tremor
5
Figure 2 From top to bottom: example of the decomposition of a pre-filtered accelerometric (ACC) signal (dashed blue line) to its intrinsic mode functions (IMFs) (full red lines) in a patient with essential tremor.
data were compared between MS patients with or without visible tremor using the Mann-Whitney test for quantitative data and the Fisher’s exact test for categorical data. The influence of gender and age between the four studied groups (MS patients with or without visible tremor, subjects with PT, and patients with ET) was studied using a
Chi2 test and Kruskal-Wallis test with post-hoc Dunn’s multiple comparison tests, respectively. The SNRs of ACC and EMG recordings were compared using Friedman’s repeated measures Anova with post-hoc Dunn’s multiple comparison tests. Finally, tremor frequencies were compared between groups using one-way Anova with Tukey-Kramer multiple
Figure 3 From top to bottom: example of the decomposition of pre-filtered accelerometric (ACC) and electromyographic (EMG) signals to their most significant intrinsic mode functions (IMFs) (left panel, 3a) and the corresponding power spectral density of each IMF, based on Welch’s algorithm (right panel, 3b), in a patient with essential tremor. EMG1 and EMG2 refer to the EMG activity recorded from the extensor and flexor carpi radialis muscles, respectively.
Please cite this article in press as: Ayache S-S, et al. Analysis of tremor in multiple sclerosis using Hilbert-Huang Transform. Neurophysiologie Clinique/Clinical Neurophysiology (2015), http://dx.doi.org/10.1016/j.neucli.2015.09.013
+Model NEUCLI-2502; No. of Pages 10
ARTICLE IN PRESS
6
S.-S. Ayache et al.
Figure 4 From top to bottom: example of pre-filtered accelerometric (ACC) and electromyographic (EMG) signals (left panel) and the corresponding Hilbert spectra of their most significant intrinsic mode functions (right panel), in a patient with essential tremor. EMG1 and EMG2 refer to the EMG activity recorded from the extensor and flexor carpi radialis muscles, respectively.
Figure 5 From top to bottom: example of the normalized Marginal Hilbert Spectra of the accelerometric (ACC) and electromyographic (EMG) signals in a patient with essential tremor. EMG1 and EMG2 refer to the EMG activity recorded from the extensor and flexor carpi radialis muscles, respectively. A clear common peak can be identified between 7 and 8 Hz.
comparisons test. A P-value of less than 0.05 was considered significant in all cases.
Results Clinical comparisons The group of MS patients with visible tremor included 12 women and 4 men, aged from 31 to 74 years (mean ± SD:
49.6 ± 10.7) with EDSS ranging from 1.5 to 8.5 (4.4 ± 2.4). The group of MS patients without visible tremor included 6 women and 4 men, aged from 35 to 58 years (47.9 ± 6.6) with EDSS ranging from 2 to 6 (3.4 ± 1.6). The group of subjects with PT included 6 women and 11 men, aged from 23 to 63 years (35.3 ± 12.4). The group of patients with ET included 2 women and 6 men, aged from 62 to 87 years (71.4 ± 8.2). In all MS patients with visible tremor at clinical examination, this tremor involved both upper limbs, was present in attitude and during movement, but was totally absent at rest. Compared to MS patients without visible tremor, cerebellar signs were more frequent (94% vs. 50%, P = 0.018, Fisher’s test) and muscle strength at the upper limb was better (18.7 ± 1.6 vs 16.9 ± 2.1, P = 0.042, Mann-Whitney test) in MS patients with visible tremor. In contrast, functional impairment and disability, as assessed by EDSS, was similar in both groups (4.4 ± 2.4 vs 3.4 ± 1.6, P = 0.508). The four studied groups differed regarding gender (P = 0.049, Chi2 test), with 75% of MS with visible tremor being women and 75% of patients with ET being men. The groups also differed regarding age (P < 0.0001, Kruskal-Wallis test), patients with ET being older than the other patients and subjects (P < 0.05, Dunn’s test).
SNR The mean SNR (dB) ranged between 28.0 ± 2.8 (MS without tremor group) and 32.7 ± 5.2 (ET group) regarding ACC recordings, between 15.6 ± 4.1 (PT group) and 20.2 ± 6.9 (MS with tremor group) regarding EMG recordings in the extensor carpi radialis muscle, and between 19.7 ± 6.6 (ET group) and 21.2 ± 8.1 (MS without tremor group) regarding EMG recordings in the flexor carpi radialis muscle. Overall, the SNR was higher for ACC recording (29.9 dB ± 6.9) than for both EMG recordings (P < 0.0001, Friedman test and P < 0.001, Dunn’s tests). Conversely, the SNR did not significantly differ between the two EMG recordings (17.9 dB ± 5.8 vs 20.6 dB ± 7.1; P > 0.05, Dunn’s test).
Please cite this article in press as: Ayache S-S, et al. Analysis of tremor in multiple sclerosis using Hilbert-Huang Transform. Neurophysiologie Clinique/Clinical Neurophysiology (2015), http://dx.doi.org/10.1016/j.neucli.2015.09.013
+Model NEUCLI-2502; No. of Pages 10
ARTICLE IN PRESS
EMD-HHT analysis of MS tremor
7
Figure 6 Individual and mean (Me) values of the highest peaks in the three normalized Marginal Hilbert Spectra of each accelerometric and electromyographic recordings in the four studied groups: subjects with physiological tremor (PT), patients with essential tremor (ET), and multiple sclerosis patients without (MS NT) or with visible tremor (MS T).
FFT-based analysis Based on FFT analysis of ACC recordings at the index finger and both EMG recordings at the forearm muscles, a common dominant frequency peak was found in the (3—8 Hz) frequency range in all patients with ET, 2/17 subjects with PT, but no patients with MS, even in the presence of visible tremor (Table 1).
EMD- and MHS-based analysis Application of procedure 1 using EMD Procedure 1 revealed the presence of common frequency peaks in the (3—8 Hz) frequency range between ACC recordings at the index finger and both EMG recordings at the forearm muscles in 10/16 MS patients with visible tremor, 4/10 MS patients without visible tremor, 5/17 subjects with PT, and all patients with ET (Table 1). The prevalence of ACC-EMG coupling did not differ between MS patients with or without visible tremor (P = 0.42, Fisher’s test). However, the frequency differed between groups (MS patients with visible tremor: 4.6 Hz ± 0.9; MS patients without visible tremor: 3.8 Hz ± 0.5; subjects with PT: 3.8 Hz ± 0.7; patients with ET: 5.8 Hz ± 1.5; P = 0.004, one-way Anova). Post-hoc Tukey-Kramer multiple comparisons test showed that tremor frequency was higher in patients with ET compared to MS patients without visible tremor and subjects with PT (P < 0.05). Application of procedure 2 using MHS By applying procedure 2, a common frequency peak in the (3—8 Hz) frequency range between ACC recordings at the index finger and both EMG recordings at the forearm muscles was found in all patients with ET, but only one MS patient with visible tremor and one subject with PT (Table 1). Application of procedure 3 for group comparisons The maximum peaks of the normalized MHSs determined in each individual and the calculated average value of these maximum peaks in the four studied groups were plotted in a graph against tremor frequency (Fig. 6). This scattering plot showed a strong overlap between values observed in MS patients with or without visible tremor and subjects
with PT. The great majority of these values were of low frequency, mainly between 3 and 5 Hz. In contrast, there was a clear distinction between these values and those observed in patients with ET.
Discussion In this work, we performed ACC and EMG recordings at the upper limb in MS patients with or without visible action tremor. EMG recordings were used to highlight rhythmic activity in the envelope of the signal and not to analyze in detail the features and the recruitment of motor unit action potentials. The quality of the measured signal is often described by the SNR value. A SNR value as low as 6 dB is sufficient for the detection of tremor burst in EMG envelope [10], although it can be difficult to find coherence in a tremor when values are below 30 dB [22]. In our study, the mean SNRs ranged between about 28 and 33 dB for ACC and 16 and 21 dB for EMG. Despite these rather low SNRs compared to other studies [29], we have been able to find common ACC-EMG rhythmic activities, demonstrating the sensitivity of our methods of analysis. It cannot however be excluded that the prevalence of these common activities was underestimated due to these low SNRs. Finally, we found similar SNR values for EMG recordings performed in the extensor carpi radialis muscle and the flexor carpi radialis muscle, although the latter muscle is situated deep underneath the skin surface in the forearm position adopted during the recordings. Using a classical FFT-based PSD analysis, we did not find any common rhythmic activity between ACC recordings at the index finger and EMG recordings at forearm muscles in MS patients, even in the presence of a visible tremor. Conversely, a common ACC-EMG rhythmic activity was found in all patients with ET and two subjects with PT. The study of the normalized MHS (procedure 2) provided similar results, revealing a common frequency peak in all patients with ET, but only one subject with PT and also one MS patient with visible tremor. In contrast to the other methods, the combined EMD-HHT analysis (procedure 1) showed a common ACC-EMG frequency peak in 54% of MS patients (62.5% in case of visible tremor and 40% in the absence of visible tremor) and 29% of subjects with PT, in addition to all patients with ET. Tremor frequency was higher in the group of patients
Please cite this article in press as: Ayache S-S, et al. Analysis of tremor in multiple sclerosis using Hilbert-Huang Transform. Neurophysiologie Clinique/Clinical Neurophysiology (2015), http://dx.doi.org/10.1016/j.neucli.2015.09.013
+Model
ARTICLE IN PRESS
NEUCLI-2502; No. of Pages 10
8
S.-S. Ayache et al. Table 1 Common frequency peaks from accelerometric and electromyographic recordings using Fast Fourier Transform (FFT), Empirical Mode Decomposition (EMD), or marginal Hilbert spectrum (MHS) analysis. Participants Common peak in MS patients with visible tremor FFT EMD 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
4.05 4.3
MHS
Common peak in MS patients without visible tremor
Common peak in subjects with PT
FFT
FFT
EMD
MHS
EMD
Common peak in patients with ET MHS
4.5 10
6.9 4.65 4 4.2
3.7 3.55 4.45
3.3
3.45 7.9
3.75 5.4 4.15 3.75 and 4.95
3.4 4.9
FFT
EMD
MHS
6.2 5.9 5.6 6.3 8.1 5.9 6.1 7.3
6.3 4.00 and 7.05 5.55 3.85 and 7.00 7.45 5.95 6.1 3.15 and 7.00
6 6 5.8 6 8 6 6 7
8
3.35 3.85
MS: multiple sclerosis; PT: physiological tremor; ET: essential tremor; ‘‘ ’’ means that no common peak was found; the values have been rounded to 0.05 or 0.10 Hz.
with ET than in the other groups, as revealed by the values of the maximum peaks of the normalized MHSs. The usual definition of tremor is purely clinical, i.e. a ‘‘rhythmical, involuntary oscillatory movement of a body part’’ [13]. Therefore, a single-channel ACC recording is sufficient to show such a ‘‘rhythmic, oscillatory activity’’ defining tremor and this approach is commonly used for the diagnosis of various types of tremor [6,23,27,47,52], including MS tremor [2]. However, to demonstrate that a pathological tremor arises from a central generator, it is crucial to show EMG synchronization at the accelerometric tremor frequency, meaning that ACC-EMG coupling drives the rhythmic movements of the tremulous limb [12,14,16,42,43]. Actually, FFT and MHS analyses failed to detect ACC-EMG coupling in almost all MS patients in our series. Conversely, EMD-HHT analysis revealed ACC-EMG coupling in 54% of MS patients, supporting the value of this procedure of signal analysis in combination with multichannel ACC-EMG recordings to reveal the presence of centrally driven pathological tremor. The second objective of this study was to compare the characteristics of MS tremor to those of ET and PT. These characteristics were very similar in MS patients (with or without visible tremor) and in subjects with PT. In both groups, we found a rather inconstant tremor at a quite low frequency, whereas patients with ET had a highly constant tremor at higher frequency. In most previously published studies, PT was studied by means of single-channel ACC recording, favoring high frequency components of tremor. Carignan et al. [5] were the first to reveal the importance of low frequency components in PT and their close
correlation with tremor amplitude. Using EMD-HHT analysis of ACC-EMG coupling, we confirmed the major impact of low-frequency components in PT. Similar low-frequency components were found in MS patients but of higher amplitude. Thus, in most MS patients, tremor could correspond to exaggerated PT, which has not been previously considered as a leading hypothesis, except in one recent study [35]. In this latter study, the authors showed that the principal ACC features of postural tremor recorded at index finger were similar in healthy individuals and MS patients, although the oscillatory output was of greater amplitude, more regular and had a higher degree of coupling in MS patients. They concluded: ‘‘this similarity suggests that, despite the neurological damage caused by this disease, the oscillators underlying tremorgenesis for the MS and healthy adults were essentially the same’’. Our results are in agreement with this conclusion. One intriguing observation was that ACC-EMG characteristics and the prevalence of ACC-EMG coupling were similar in the two groups of MS patients whether tremor was ‘‘visible’’ or not. In fact, MS patients with ‘‘visible tremor’’ only differed from MS patients without ‘‘visible tremor’’ by higher prevalence of cerebellar signs and greater muscle strength at the upper limb. Thus, MS tremor could be a form of PT whose ‘‘visibility’’ is enhanced by cerebellar dysfunction in the absence of motor deficit. Conversely, the presence of motor weakness could reduce the visibility of PT. In fact, PT is a ubiquitous property of the neuromuscular system that becomes exaggerated and more visible in certain conditions like fatigue, stress, and hypothyroidism, among others. Cerebellar dysfunction in MS patients could
Please cite this article in press as: Ayache S-S, et al. Analysis of tremor in multiple sclerosis using Hilbert-Huang Transform. Neurophysiologie Clinique/Clinical Neurophysiology (2015), http://dx.doi.org/10.1016/j.neucli.2015.09.013
+Model NEUCLI-2502; No. of Pages 10
ARTICLE IN PRESS
EMD-HHT analysis of MS tremor be one of these conditions, since cerebellum plays a key role in the control of movement. In addition, in the case of cerebellar dysfunction, moving a limb to an intended target or maintaining it in a given position leads to continuous adjustments and readjustments, which can be mistaken for tremor. It is likely that cases of ‘‘visible tremor’’ noticed in MS patients could in fact reveal the impact of dysmetria or cerebellar ataxia. In a previous report, Sabra and Hallett [46] expressed the same opinion that ‘‘dissecting intention tremor from serial dysmetria and postural tremor from ataxia is difficult’’. The term ‘‘pseudotremor’’ covers pathological conditions such as dysmetria and ataxia that produce pseudorhythmic movements mimicking tremor in patients with neurological disorders. The high prevalence of cerebellar alterations in MS patients may be a primary cause of tremor-like activity in these patients [1,45,50].
Conclusion To summarize, our results suggest that ‘‘visible tremor’’ in MS patients could correspond to an exaggerated PT (with ACC-EMG coupling) or a pseudorhythmic activity (‘‘pseudotremor’’) related to dysmetria or ataxia (without ACC-EMG coupling). Both conditions are favored by cerebellar dysfunction, but cannot be considered strictly speaking as forms of ‘‘pathological tremor’’, directly generated by a central oscillator that would be originally created by MS lesions. We cannot rule out however the possible contribution of a low-frequency tremor originating from lesions in the brainstem (midbrain) or cerebellothalamic circuits, but such a ‘‘pathological tremor’’ may be much less frequent than expected in MS. The limitations of this study were a small sample size of heterogeneous MS patients and the absence of neuroimaging assessment. Therefore, further studies should be designed to confirm these results in a larger population of MS patients and to correlate ACC-EMG features to the location of lesions or metabolic changes into the brain. On clinical examination, it is difficult to characterize MS tremor perfectly. Multichannel ACC-EMG recordings with EMD-HHT analysis may help to differentiate which is related to an exaggerated PT, a ‘‘pseudotremor’’, or a ‘‘pathological tremor’’. This better characterization might be important to distinguish different underlying pathophysiological mechanisms that can respond differentially to treatment, as previously suggested by the predictive value of tremor analysis in MS patients treated by thalamotomy [32]. Finally, the real value of our EMD-HHT approach is also to be able to reveal surface EMG-ACC coupling when the SNR of these recordings is low.
Disclosure of interest The authors declare that they have no competing interest.
References [1] Alusi SH, Glickman S, Aziz TZ, Bain PG. Tremor in multiple sclerosis. J Neurol Neurosurg Psychiatry 1999;66:131—4. [2] Alusi SH, Worthington J, Glickman S, Bain PG. A study of tremor in multiple sclerosis. Brain 2001;124:720—30.
9 [3] Ayache SS, Al-Ani T, Lefaucheur JP. Distinction between essential and physiological tremor using Hilbert-Huang Transform (HHT). Neurophysiol Clin 2014;44:203—12. [4] Barreca SR, Stratford PW, Lambert CL, Masters LM, Streiner DL. Test-retest reliability, validity, and sensitivity of the Chedoke arm and hand activity inventory: a new measure of upperlimb function for survivors of stroke. Arch Phys Med Rehabil 2005;86:1616—22. [5] Carignan B, Daneault JF, Duval C. Quantifying the importance of high frequency components on the amplitude of physiological tremor. Exp Brain Res 2010;202:299—306. [6] Costa J, González HA, Valldeoriola F, Gaig C, Tolosa E, Valls-Solé J. Nonlinear dynamic analysis of oscillatory repetitive movements in Parkinson’s disease and essential tremor. Mov Disord 2010;25:2577—86. [7] De Luca CJ, Contessa P. Hierarchical control of motor units in voluntary contractions. J Neurophysiol 2012;107:178—95. [8] De Luca CJ, Gilmore LD, Kuznetsov M, Roy SH. Filtering the surface EMG signal: movement artifact and baseline noise contamination. J Biomech 2010;43:1573—9. [9] De Luca CJ, Hostage EC. Relationship between firing rate and recruitment threshold of motoneurons in voluntary isometric contractions. J Neurophysiol 2010;104:1034—46. [10] De Marchis C, Conforto S, Severini G, Schmid M, D’Alessio T. Detection of tremor bursts from the sEMG signal: an optimization procedure for different detection methods. Conf Proc IEEE Eng Med Biol Soc 2011;2011:7508—11. [11] Deuschl G, Lauk M, Timmer J. Tremor classification and tremor time series analysis. Chaos 1995;5:48—51. [12] Deuschl G, Krack P, Lauk M, Timmer J. Clinical neurophysiology of tremor. J Clin Neurophysiol 1996;13:110—21. [13] Deuschl G, Bain P, Brin M, Ad Hoc Scientific Committee. Consensus statement of the Movement Disorder Society on tremor. Mov Disord 1998;13(Suppl. 3):2—23. [14] Deuschl G, Raethjen J, Lindemann M, Krack P. The pathophysiology of tremor. Muscle Nerve 2001;24:716—35. [15] Donnelly D. The Fast Fourier and Hilbert-Huang Transforms: a comparison. IJCCC 2006;1:45—52. [16] Elble RJ. Characteristics of physiologic tremor in young and elderly adults. Clin Neurophysiol 2003;114:624—35. [17] Elble RJ, Randall JE. Motor-unit activity responsible for 8- to 12-Hz component of human physiological finger tremor. J Neurophysiol 1976;39:370—83. [18] Elble RJ, Randall JE. Mechanistic components of normal hand tremor. Electroencephalogr Clin Neurophysiol 1978;44: 72—82. [19] Farina D, Merletti R, Enoka RM. The extraction of neural strategies from the surface EMG: an update. J Appl Physiol (1985) 2014;117:1215—30. [20] Fox JR, Randall JE. Relationship between forearm tremor and the biceps electromyogram. J Appl Physiol 1970;29: 103—8. [21] Gironell A, Kulisevsky J, Pascual-Sedano B, Barbanoj M. Routine neurophysiologic tremor analysis as a diagnostic tool for essential tremor: a prospective study. J Clin Neurophysiol 2004;21:446—50. [22] Hellwig B, Häussler S, Schelter B, Lauk M, Guschlbauer B, Timmer J, et al. Tremor-correlated cortical activity in essential tremor. Lancet 2001;357:519—23. [23] Hossen A, Muthuraman M, Al-Hakim Z, Raethjen J, Deuschl G, Heute U. Discrimination of Parkinsonian tremor from essential tremor using statistical signal characterization of the spectrum of accelerometer signal. Biomed Mater Eng 2013;23: 513—31. [24] Huang NE. Introduction to Hilbert-Huang Transform and its related mathematical problems. In: Huang NE, Shen SSP, editors. Hilbert-Huang Transform and its applications. Singapore: World Scientific; 2005. p. 1—26.
Please cite this article in press as: Ayache S-S, et al. Analysis of tremor in multiple sclerosis using Hilbert-Huang Transform. Neurophysiologie Clinique/Clinical Neurophysiology (2015), http://dx.doi.org/10.1016/j.neucli.2015.09.013
+Model NEUCLI-2502; No. of Pages 10
ARTICLE IN PRESS
10 [25] Huang NE, Wu Z. A review on Hilbert-Huang Transform: method and its applications to geophysical studies. Rev Geophys 2008;46:RG2006. [26] Huang NE, Shen Z, Long SR, Wu MC, Shih HH, Zheng Q, et al. The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis. Proc R Soc London 1998;454:903—95. [27] Jang W, Han J, Park J, Kim JS, Cho JW, Koh SB, et al. Waveform analysis of tremor may help to differentiate Parkinson’s disease from drug-induced parkinsonism. Physiol Meas 2013;34:N15—24. [28] Koch M, Mostert J, Heersema D, De Keyser J. Tremor in multiple sclerosis. J Neurol 2007;254:133—45. [29] Kugler P, Jaremenko C, Schlachetzki J, Winkler J, Klucken J, Eskofier B. Automatic recognition of Parkinson’s disease using surface electromyography during standardized gait tests. Conf Proc IEEE Eng Med Biol Soc 2013;2013:5781—4. [30] Kurtzke JF. Rating neurologic impairment in multiple sclerosis: an expanded disability status scale (EDSS). Neurology 1983;33:1444—52. [31] Li K, Hogrel JY, Duchêne J, Hewson DJ. Analysis of fatigue and tremor during sustained maximal grip contractions using Hilbert-Huang Transformation. Med Eng Phys 2012;34: 832—40. [32] Liu X, Aziz TZ, Miall RC, Rowe J, Alusi SH, Bain PG, et al. Frequency analysis of involuntary movements during wrist tracking: a way to identify ms patients with tremor who benefit from thalamotomy. Stereotact Funct Neurosurg 2000;74:53—62. [33] Medical Research Council. Aids to examination of the peripheral nervous system. In: Memorandum no. 45. London: Her Majesty’s Stationary Office; 1976. [34] Mitra PP, Pesaran B. Analysis of dynamic brain imaging data. Biophys J 1999;76:691—708. [35] Morrison S, Sosnoff JJ, Sandroff BM, Pula JH, Motl RW. The dynamics of finger tremor in multiple sclerosis is affected by whole body position. J Neurol Sci 2013;324:84—9. [36] Muthuraman M, Hossen A, Heute U, Deuschl G, Raethjen J. A new diagnostic test to distinguish tremulous Parkinson’s disease from advanced essential tremor. Mov Disord 2011;26: 1548—52. [37] Myers LJ, Lowery M, O’Malley M, Vaughan CL, Heneghan C, St Clair Gibson A, et al. Rectification and non-linear preprocessing of EMG signals for corticomuscular analysis. J Neurosci Methods 2003;124:157—65. [38] O’Suilleabhain PE, Matsumoto JY. Time-frequency analysis of tremors. Brain 1998;121:2127—34.
S.-S. Ayache et al. [39] Palit SK, Mukherjee S, Bhattacharya DK. New types of nonlinear autocorrelations of bivariate data and their applications. Appl Math Comput 2012;218:8951—67. [40] Pittock SJ, McClelland RL, Mayr WT, Rodriguez M, Matsumoto JY. Prevalence of tremor in multiple sclerosis and associated disability in the Olmsted County population. Mov Disord 2004;19:1482—5. [41] Polman CH, Reingold SC, Banwell B, Clanet M, Cohen JA, Filippi M, et al. Diagnostic criteria for multiple sclerosis: 2010 revisions to the McDonald criteria. Ann Neurol 2011;69: 292—302. [42] Raethjen J, Pawlas F, Lindemann M, Wenzelburger R, Deuschl G. Determinants of physiologic tremor in a large normal population. Clin Neurophysiol 2000;111:1825—37. [43] Raethjen J, Lauk M, Köster B, Fietzek U, Friege L, Timmer J, et al. Tremor analysis in two normal cohorts. Clin Neurophysiol 2004;115:2151—6. [44] Roy SH, Cheng MS, Chang SS, Moore J, De Luca G, Nawab SH, et al. A combined sEMG and accelerometer system for monitoring functional activity in stroke. IEEE Trans Neural Syst Rehabil Eng 2009;17:585—94. [45] Ruutiainen J. Assessment and treatment of ataxia in multiple sclerosis. In: Ketelaer P, Prosiegel M, Battaglia M, Messmer U, editors. A problem-oriented approach to multiple sclerosis. Leuven: Amersfoort; 1997. p. 227—35. [46] Sabra AF, Hallett M. Action tremor with alternating activity in antagonist muscles. Neurology 1984;34:151—6. [47] Shaikh AG, Jinnah HA, Tripp RM, Optican LM, Ramat S, Lenz FA, et al. Irregularity distinguishes limb tremor in cervical dystonia from essential tremor. J Neurol Neurosurg Psychiatry 2008;79:187—9. [48] Timmer J, Haussler S, Lauk M, Lucking CH. Pathological tremors: deterministic chaos or nonlinear stochastic oscillators? Chaos 2000;10:278—88. [49] Veltcheva A, Guedes SC. Identification of the components of wave spectra by the Hilbert Huang Transform method. Appl Ocean Res 2004;26:1—12. [50] Weinshenker BG, Issa M, Baskerville J. Long-term and shortterm outcome of multiple sclerosis: a 3-year follow-up study. Arch Neurol 1996;53:353—8. [51] Welch P. The use of Fast Fourier Transform for the estimation of power spectra: a method based on time averaging over short, modified periodograms. IEEE Trans Audio Electroacoustics 1967;15:70—3. [52] Zeuner KE, Shoge RO, Goldstein SR, Dambrosia JM, Hallett M. Accelerometry to distinguish psychogenic from essential or parkinsonian tremor. Neurology 2003;61:548—50.
Please cite this article in press as: Ayache S-S, et al. Analysis of tremor in multiple sclerosis using Hilbert-Huang Transform. Neurophysiologie Clinique/Clinical Neurophysiology (2015), http://dx.doi.org/10.1016/j.neucli.2015.09.013