A novel method for the quantification of key components of manual dexterity after stroke

Subjects

Ten adult stroke patients were recruited at the Rehabilitation clinic at Sainte-Anne
Hospital, Paris. All patients suffered from a single ischemic or hemorrhagic stroke
and were at least 2 weeks post-stroke at the time of their participation to the study.
Included patients had varying degrees of hemiplegia affecting the upper limb and the
hand. Exclusion criteria comprised severe loss of sensation of the affected limb,
other neurological conditions and cognitive dysfunction that would interfere with
the understanding of the experiment, such as visual deficits or severe neglect. Ten
healthy control subjects, comparable in age, were also recruited. Table 1 lists the demographic and clinical details. The procedures of the study complied
with the Declaration of Helsinki, and subjects provided informed consent.

Table 1. Clinical measures

Clinical measures

The Arm Research Action Test (ARAT), a clinical test for grasp, grip, pinch and gross
movement in the hemiparetic hand, was used as a global measure of hand function 33], 34]. The Moberg pick-up test was used as a clinical assessment of grip function in each
hand. Time taken to place all 12 objects into the box was recorded. The time taken
reflects the degree of precision grip function (18 s is considered pathological in
this age span) 35]. A Semmes-Weinstein mono-filament test with three calibers (2 g, 0.4 g and 0.07 g)
was used to measure the tactile sensitivity of finger tips in each hand 36]. Maximal grip force (in Kg) in each hand was recorded (best of two trials) with a
hydraulic Jamar dynamometer (http://www.lafayetteevaluation.com). Proprioception was tested by assessing the subjects’ capacity to detect and match
passive finger displacement in one hand while keeping the eyes shut and rated as intact,
impaired or absent. All measures were also obtained in control subjects, except the
ARAT.

Finger Force Manipulandum (FFM)

Together with Sensix (www.sensix.fr) we developed the Finger Force Manipulandum (FFM) in order to quantify key components
of manual dexterity in stroke (and other) patients. The FFM is equipped with four
pistons positioned under the tip of the index, middle, ring and little finger, each
coupled to an individual strain gauge force sensor (Fig. 1). The height of the pistons can be adjusted but in this study we used a constant
piston height of 15 mm across all subjects. Pistons have a contact surface of 15 mm
diameter and are 20 mm apart. With increasing force the pistons move against a spring
load over a range of 10 mm. The end of this dynamic (non- static) range is reached
with 1N. Above 1N, forces are controlled isometrically. Thus each sensor measures
force along the piston axis exerted from each finger independently. The precision
of the sensor is 0.01N, with a range of 0–9N. Force data of each finger was sampled
to a CED 1401 (with 10 kHz sampling rate/digit) connected to a computer running Spike
2v6 (Cambridge Electronic Design, www.ced.co.uk) software. Custom-written CED scripts provided real-time visual display of digit
forces and target instructions or target forces.

Fig. 1. The Finger Force Manipulandum (FFM). Index, middle, ring and little finger each apply
forces on a spring-loaded piston. Two types of tasks were implemented: continuous
force tracking and finger tapping. Forces applied by each finger were recorded via
a CED interface (not shown) and used for real-time visual feedback and for performance
analysis

FFM tasks

Four separate tasks (i-iv) were developed in order to quantify different components
of manual dexterity. The finger force tracking task was developed in order to measure
the capacity to generate and control fingertip forces 18]. The sequential finger tapping task was developed in order to assess the ability
to learn and recall finger movement sequences 37]. The single finger tapping task is a timing task designed to test the capacity to
perform repetitive tapping with and without auditory cues 9]. The multi-finger tapping task was designed to test the independence of finger movements
in one-finger configurations 22], 38] and two-finger configurations. Each of the four tasks comprised different conditions
in order to evaluate performance across varying forces, tapping frequencies, and fingers.
In all tasks the subject was first required to place the fingers on the pistons and
was instructed to maintain the fingers on the pistons throughout the tasks. Every
subject was able to use the FFM with the forearm supported on the table and the shoulder
was in a relaxed slightly flexed position. To ensure a comfortable position some subjects
used a silicone wrist support during the tasks.

(i) The Finger Force–Tracking task is a visuo-motor task of finger force control. By varying the force on the piston
with the finger, the subject controlled a cursor on a computer screen (Fig. 2a). The subject was instructed to follow the target force as closely as possible with
the cursor. The target force (a line) passed from right to left over the screen, presenting
successive trials. Each trial consisted of a ramp phase (a linear increase of force
over a 1.5 s period), a hold phase (a stable force for 4 s) and a release phase (an
instantaneous return to the resting force level, 0N) followed by a resting phase (2 s).
Trials were repeated 24 times, distributed in four blocks of 6 trials, two blocks
with a target force of 1N and two with a target force of 2N. These low absolute forces
were chosen since dexterous action usually employs low forces at which key sensory
events occur 39]. In this study, patients performed the finger force-tracking task separately with
the index and the middle finger of their hemiparetic hand and controls performed the
task with their index and middle finger of their right hand. Task duration was 3 min
20 s/digit.

Fig. 2. The four FFM tasks. a–d: Left panels: Setup with FFM and screen providing visuo-motor feedback. Right panels:
Example recordings of finger force traces. Index finger: red, middle: blue, ring:
green, little: turquoise. The target for each finger is shown as a line of the same
color (trapezoid form in a, b, d). Left column: control subject. Right column: stroke patient. aFinger force tracking. Screen: The yellow line represents the target force and the cursor (here close to
the ramp) represents the instantaneous force exerted by the index finger. The subject
has to match the vertical cursor position with the target force. Right panels: tracking
examples of five successive trials. Note: the patient’s tracking force trace is more
irregular, does not return to baseline between trials and the little finger (turquoise)
applies unwanted force (motor overflow). bSequential finger tapping: Screen: the height of 4 red vertical bars represents the force exerted by each finger.
Next to each finger feedback the target bar (white), here only visible for the index
finger. Successively appearing target bars indicate the 5-tap finger sequence (e.g.,
digit 3-2-4-5-3). Right panels: correct tapping sequence for the control subject,
erroneous sequence in the patient. cSingle finger tapping: Screen: ring finger is indicated as tapping finger (white bar). Visual feedback
was only provided for the tapping finger (red bar). Right: index finger 1Hz condition
with (15 s) and without (20s) tapping cue. Less finger taps, incomplete return to
baseline and unwanted movements of other fingers are noticeable in the patient. d) Multi–finger tapping: Screen: two-finger target tap (index and ring finger, white bars) and corresponding
two-finger user tap (red bars). Right: four subsequent trials, each with a different
finger combination (ring-little; little; middle-ring; index). The patient clearly
has more difficulties

(ii) The Sequential finger tapping task is a 5-tap finger sequence involving the four digits. The visual display consisted
of 4 columns (representing the 4 digits), whose height varied in real-time as a function
of exerted finger force (feedback). In addition, a target column (cue) adjacent to
each feedback column indicated the piston to be pressed (Fig. 2b). The subject was instructed to press the indicated piston as soon as the target
appeared. The 5 successive targets of a given sequence appeared at a rate of 1 Hz.
Each sequence was repeated 10 times with visual cues (learning phase) and then repeated
5 times from memory, i.e. without cues, and as quickly as possible (recall phase).
Force feedback was always present. Subjects were instructed to match the tap force
approximately to target of 2N (same for the other tapping tasks). In this protocol,
the subjects performed three previously unknown motor sequences: they first learned
and then repeated the sequence (A) 2-5-3-4-2 (2?=?index; 5?=?little); then the sequence
(B) 4-3-5-2-4 and finally the sequence (C) 3-2-4-5-3. A single sequence (trial) of
5 taps lasted 5 s and the duration for all 15 trials was 2 min 20 s.

(iii) The Single finger tapping task consisted of repetitive tapping with one finger with or without an auditory
cue. The visual display was similar to that in task (ii) and indicated which finger
to tap but did not provide any timing cue. Three tapping rates were tested: 1, 2 and
3Hz (similar to 9]). After the cued tapping period (15 taps) the subject was required to continue tapping
for a similar period, without cue but at the same rate. The subject started at 1 Hz
with the index finger, followed by the middle (Fig. 2c), ring and little finger. This protocol was repeated at 2 Hz and then at 3 Hz. The
total duration of this task was 4 min.

(iv) The Multi–finger tapping task consisted of simultaneous tapping with different finger configurations in response
to visual instructions. The visual display was similar to that in task (ii) and (iii).
Subjects were instructed to reproduce 11 different finger tap configurations following
the visual cue (Fig. 2d). The 11 different configurations consisted of 4 one-finger taps (separate tap of
index, middle, ring or little finger), 6 two-finger configurations (simultaneous index-middle,
index-ring, index-little, middle-ring, middle-little or ring-little finger taps),
and one four-finger tap. All configurations were performed twice resulting in a total
of 32 (4 × 8) one-finger taps, 30 (6 × 5) two-finger taps and 2 four-finger taps.
Performance measures were calculated for one and two-finger configurations. Four finger
taps were not analyzed. The order of the configurations was pseudo-randomized with
equal number of transitions between one and two-finger taps. The entire task with
its 64 trials lasted 4 min and 40 s.

Data analysis

Task performance was analyzed using Matlab (v7.5, The MathWorks, Inc., Natick, MA,
USA). The four force signals were first down-sampled to 100 Hz for the analysis.

Finger force tracking: all performance measures were calculated trial-by-trial (N?=?24). Tracking error
was calculated as the root-mean-square error (RMSE) between the actual applied force
and the target force. The error was separately extracted during the ramp and the hold
phase. The time of the force onset in response to the target ramp and the time of
the release onset at the end of the hold phase were calculated as threshold crossings
of dF/dt. The release duration was computed as the time taken to reduce the force
from 75 to 25 % of the target force 18]. The coefficient of variation (CV) of force (i.e. SD/mean across time bins) was calculated
during the hold phase and averaged across trials. Mean force during the hold was calculated
as the average force across 3 s excluding the first and last 500 ms of the hold phase.
Mean baseline force was calculated as the average force during the resting phase between
each trial from 1500 ms to 500 ms before the ramp onset.

For the three tapping tasks the finger taps were identified in a similar way. Starting from the force trace
each tap was identified as a discrete event according to threshold (0.5N) allowing
identification of target and the applied force peaks (retained as taps). The time
location and amplitude of each tap were then recorded. Subsequently, the following
task-specific performance variables were obtained:

In the Sequential finger tapping task we computed the number of user taps trial-by-trial, i.e. for each 5-tap target
sequence. By comparing the user taps to the target sequence, each trial was then labeled
as correct or incorrect. In case of an incorrect sequence the number of missing or
additional unwanted taps was recorded, as well as the number of consecutive correct
taps within parts of the sequence. Furthermore, performance was calculated across
trials, by computing the number of correct trials and the number of error taps for
each finger. These measures were obtained for the learning and the recall phase, respectively.

In the single finger tapping task the lead-finger (target finger) and the non-lead-fingers were identified in
each condition (finger and 1, 2 or 3 Hz). For the lead-finger the number of taps,
the tap amplitude, and the interval between consecutive taps were calculated for each
condition. Unwanted taps were identified in the non-lead-fingers and labeled as overflow
taps (non-lead-finger tap at the same time as a lead-finger tap) or as unwanted finger
taps (non-lead-finger tap in the absence of a lead-finger tap). To estimate the capability
to adapt the tapping rate to the target frequency of the cue we calculated the slope
of the tapping rate across the 1 Hz, 2 Hz and 3 Hz conditions. A slope?=?1 indicates
correct tapping rate, a slope??1 slower execution.

In the multi–finger tapping task each tap, in response to a displayed finger configuration, was identified as
correct or incorrect (success rate), i.e. identical to or different from the required
target taps. Errors, in each finger, were categorized as missing taps (omissions,
omission rate), or as unwanted extra-finger-taps (one or several) (similar to errors
reported in keyboard typing 40]). Across trials the number of errors was evaluated as a function of the target (one-
or two-) finger configuration.

Finally, in order to obtain individual profiles of components of manual dexterity,
we plotted each patient’s performance in the six most discriminatory variables (showing
group differences) and compared it to the performance range observed in the control
group. Values beyond the control group’s mean?+?2SD in a given measure were considered
indicative of pathological performance.

Statistical analysis

Descriptive statistics are shown as mean?±?SD. Student’s T-test was used to test for group differences in single-level variables. Differences
in the measures obtained from the four tasks described above were tested using repeated
measures ANOVAs. (i) Force tracking: independent variables (error, timing, etc.) were
studied with ANOVA including one between-group factor GROUP (patients, controls),
and three within-subject levels: FINGER (index, middle), FORCE (1N, 2N), PHASE (Ramp,
Hold). (ii) Sequential finger tapping: independent variables (success rate, number
of correct taps) were studied with ANOVA including one between-group factor GROUP
(patients, controls), and two within-subject levels: SEQUENCE (sequence A, B, C),
PHASE (learning and recall phase). (iii) Single finger tapping: independent variables
(tapping rate, number of overflow taps, etc.) were studied with ANOVA including one
between-group factor GROUP (patients, controls), and three within-subject levels:
FREQUENCY (1, 2, 3 Hz), FINGER (index, middle, ring, little) and PHASE (with auditory
cue, without auditory cue). (iv) Multi-finger tapping: independent variables (success
rate, number of unwanted extra finger taps, etc.) were studied with ANOVA including
one between-group factor GROUP (patients, controls). Post-hoc tests were performed
using Fisher LSD Test. Spearman’s rank order correlation was used to investigate correlations
between performance measures and clinical scores. Jamar and Moberg Pick up scores
were presented as % of non-hemiparetic hand scores for correlation tests. Pearson’s
correlation was used to test for relations between different performance measures.
The level of significance was set to p??0.05.