Increased task-uncorrelated muscle activity in childhood dystonia

Participants

Inclusion criteria for this study were: I) primary or secondary dystonia; II) pediatric
age (8-21 years); III) upper limb control impairment that does not prevent the writing
task execution; IV) no cognitive impairment that prevents understanding of instructions;
V) no Deep Brain Stimulation. Participants (Table 1) consisted of 7 children with dystonia (2 girls, 5 boys; ages 8–19 years, mean 12.6?±?4.7 years)
recruited from the Children’s Hospital Los Angeles and diagnosed by a pediatric neurologist,
and a control group with 9 healthy children (7 girls, 2 boys; ages 12–20 years, mean
15.8?±?4 years). The age distributions of the two groups were not statistically different.
All participants with dystonia were rated on the Barry-Albright Dystonia (BAD) Scale
10] by two raters and the average score is reported in Table 1. The University of Southern California Institutional Review Board approved the study
protocol. All parents gave informed written consent for participation and authorization
for use of protected health information, and all children gave written assent. The
study was performed in accordance with the Declaration of Helsinki.

Table 1. Characteristics of subjects

Apparatus

The acquisition system synchronized upper limb kinematics (Flock of Birds®, Ascension,
Burlington, VT USA; 100 Hz sample frequency), EMG (DataLOG MWX8, Biometrics Ltd, Newport,
UK; 1000 Hz sample frequency; 20-460 Hz bandwidth), and 2D coordinates of the pen
tip on a tablet (iPad®, Apple®, Cupertino, CA USA; 60 Hz sample frequency). Kinematic
sensors were placed on elbow lateral epicondyle, elbow medial epicondyle, and acromion
of the upper-limb used to perform the task. Elbow coordinates were computed as the
middle point between the two sensors on the elbow. For all the subjects, surface EMG
signals were extracted from eight muscles of the upper limb that are known to contribute
to wrist, elbow, and shoulder movements: Flexor Carpi Ulnaris (FCU), Extensor Carpi
Radialis (ECR), Biceps Brachii (BIC), Triceps Brachii (TRIC), Anterior Deltoid (AD),
Lateral Deltoid (LD), Posterior Deltoid (PD), and Supraspinatus (SS) (Table 2). Eight bipolar surface active electrodes (SX230 from Biometrics Ltd, Newport, UK),
with an inter-electrode distance of 20 mm, were placed on the subjects. Prior to the
placement, the skin over the muscles and the surface of the sensors were wiped with
isopropyl alcohol pads to reduce electrical impedance at the skin electrode interface.
The target muscles were mostly found by palpation, anatomical landmarks 11], and by visual inspection of the signal that gave the best response to clinical tests
reported in Table 2. A single electrode was also placed on the subject’s contralateral wrist to serve
as a ground and reference electrode. Prior to the start of the experiment, the EMG
signals were displayed on a real-time monitor and visually inspected to ensure proper
placement and quality of the signal. Custom software on the tablet was developed to
record the 2D coordinates of the pen tip on the tablet (Cocos2d development environment;
iOS 4.3 operating system; Apple®, Cupertino, CA USA). The subjects were seated on
an armless chair, positioned at a distance from a height-adjustable table that allowed
them to reach the furthest point of the tablet (fixed on the table in portrait orientation)
with the elbow at 90 % of its maximum extension (Fig. 1). The subjects’ trunk was fastened to the seatback with a Velcro® belt to prevent
the subjects from bending the trunk towards the table.

Table 2. Target muscles with related functions and clinical tests

Fig. 1. Setup. During the execution of the figure-eight writing task, subjects were seated
at a height-adjustable desk. The apparatus included a motion tracking system (upper
limb kinematics), an electromyography device (surface EMG of eight upper limb muscles),
and a tablet (2D coordinates of the pen tip)

Protocols

Motor performance was studied during the execution of figure-eight writing movements.
The subjects were asked to draw a figure-eight (15.7 (Y) cm × 7.8 (X) cm) on the tablet,
following a displayed thin trace (0.3 cm thick). Prior to the start of the experiment,
participants were encouraged to be as accurate as possible while tracing the figure-eight
at their natural speed. Starting from the upper point of the figure-eight, subjects
were requested to move in the mediolateral direction opposite to the arm used to perform
the task. Subjects drew three sequences of ten figure-eight movements in a row. The
task was performed with the dominant or preferred arm and subjects were asked not
to lean their forearm on the table while writing.

Data analyses

Data analyses were executed with Matlab® R2011a software (Mathworks®, Natick, MA USA).
Statistical analysis was performed using RStudio® Version 0.98.981 (RStudio Inc.^©, Boston, MA, USA).

Joint EMG-kinematic analysis

Before performing the spectral analysis, each EMG signal was processed with a band-pass
Butterworth filter (5^th order, 5-400 Hz), and a stop-band Butterworth filter (5^th order, 60 Hz). A nonlinear recursive filter based on Bayesian estimation was applied.
Compared to the traditional linear amplitude envelope, Bayesian filtering produces
a smooth output that estimates the driving force underlying the EMG signal with low
variability yet with the possibility of very rapid changes in output 12]. This filtering is essential to remove variability due to the surface EMG itself,
so that all remaining variability is due to the neural control of muscle activity.
Signals were then normalized to the maximum activation levels during movement, thus
obtaining signals ranging from 0 to 1.

In order to detect the frequency features related to the motor outcome on the EMG
signals, spectral analysis was applied to kinematic (Y_tablet and X_tablet) and EMG data (normalized Bayesian filtering outputs). Each kinematic and EMG signal
was pre-processed as follows: I) the sequence was divided into the 10 single figure-eight
movements; II) each kinematic and EMG movement was re-sampled to equalize the duration
of the figure-eight movements between all subjects; III) re-sampled movements were
re-assembled in order to rebuild the sequence; IV) the signal was linearly de-trended;
V) Fourier Transform (FT) of the re-sampled sequence was computed. We then computed
the Power Spectral Density (PSD) based on the FT coefficients for kinematic and EMG
signals. In the figure-eight, the horizontal (f_x) and vertical (f_y) frequency components are expected to be in a ratio of 2:1 (f_x?=?2 * f_y). As a result, for each subject, the Y_tablet PSD presented a well-defined peak at the frequency related to the mean duration of
the figure-eight movement (f_y), while the X_tablet PSD showed a peak at double the figure-eight frequency (f_x). Spectral peaks at exactly the same frequencies, f_y and f_x, were clearly detectable also in the EMG PSDs (Fig. 2). The sum of the spectral energy of the peaks at f_y and f_x (P_y?
+?P_x) was regarded as an index of muscle activity that contains the frequency components
of the kinematic task, and it was computed for each muscle individually. All PSD components
at frequencies other than f_y and f_x were considered task-uncorrelated, representing noisy and variable components that
do not contribute to the desired cyclical task, together with corrective activity
to counter possible errors generated while non-efficiently tracing the figure-eight.
The more noise that characterizes the EMGs, the more we expected the task-uncorrelated
components to increase compared with the task-correlated components. The ratio between
P_y?
+?P_x and the full spectrum energy ranging from 0 to 5 Hz was calculated for each muscle
as an indication of the relative contribution of the task-correlated components (task-correlation index) (Table 3). For each subject, for each muscle, the task-correlation index was computed for each sequence of ten single figure-eight movements and averaged
over the three sequences performed.

Fig. 2. EMG-Kinematics Spectral Analysis. Panel a: Control subject (c2); Panel b: Subject with dystonia (d2). For each panel, from top to bottom: Tablet y-trajectory,
Tablet x-trajectory, Triceps Brachii (TRIC) and Posterior Deltoid (PD) EMGs and non-linear
envelopes (Filt) in a sequence of ten figure-eight movements represented in time (left
column) and frequency (right column) domains. Note that Filt signals are normalized
and dimensionless both in time and frequency domains (n.u.). f_y and f_x represent the subject-specific frequencies related to the vertical and horizontal
components of the figure-eight

Table 3. Task-correlation index

Kinematic analysis

Kinematic data were processed with a low-pass Butterworth filter (5^th order, 3 Hz). The cutoff frequency was determined by the “Jackson Knee” method 13].

The accuracy error of the figure-eight trace on the tablet was computed as the root mean square error
between the actual pen trajectory and the displayed figure-eight trace for each movement
separately, and the value was then averaged over all movements. The speed was computed for each figure-eight movement and then averaged over the thirty single
movements (three sequences of ten movements). Moving fast and accurately can be considered
a central goal of the motor performance. Since motor speed and accuracy interact,
to quantify motor performance, it is important to measure them together 7], 14]–16]. Therefore, we examined motor skill during the writing task by computing the ratio
between accuracy error and speed (AccErr/Speed) for each movement separately and we averaged it over the thirty movements. The intra-subject
spatial variability was estimated as the average standard deviation of the trajectories after time alignment.
As proposed in 17], time alignment was achieved by phase shifting the data in the frequency domain.
In particular, after re-sampling all the figure-eight movements separately to equalize
their duration, the FT of the module of the position vector of each figure-eight movement
was computed and all the FT components of the i-th figure-eight movement were shifted by ?(f_yi), where ?(f_yi) is the phase of the frequency component related to the duration of the i-th figure-eight movement (f_yi). The signals were then reconstructed by applying an inverse FT. As a result, the
different movements were aligned in time and the variability was determined only by
spatial characteristics. Spatial variability was evaluated for tablet, elbow, and acromion trajectories. The intra-subject temporal variability was quantified as the standard deviation among the durations of each movement.

Statistical analysis

Due to the small sample size, we ran nonparametric statistics. Nonparametric aligned
rank test for interaction in two-way factorial designs with repeated measures (R-package
‘npIntFactRep’) was applied to investigate a possible between-group difference in
the task-correlation index. The model included a between factor with 2 levels (Group) and a within factor with
8 levels (Muscle). For the kinematic parameters, to look for possible between-group
differences, we used the nonparametric Kruskal-Wallis rank sum test. Spearman’s rank
correlation coefficient was used to investigate the presence of statistical dependence
between the task-correlation index and all the kinematic parameters separately. For children with dystonia, the Spearman’s
rank correlation coefficient was also studied between the task-correlation index and the BAD score of the arm used to perform the task (BAD_arm), and the BAD total score (BAD_total) (Table 1). For all tests, the significance level was set at 5 %.