212
U. Wijk et al.
If the subject did not respond within 10 s, the next stimulation
begun automatically and a non-answer was recorded.
Stimulation set-up
The study setup consisted of a tactile display (28) using 5
HS-40 Nano analogue servo motors (HI-TEC RCD, USA)
incorporated in a silicone cuff with 3D-printed boxes. The
boxes were placed in an upside-down U-shape, resembling the
positions of the fingertips (Fig. 1), similar to previous work
(28). During the sessions, the tactile display was placed on the
left forearm. When positioning the tactile display for the first
time, the boundaries were marked on the skin of the subject to
ensure identical placement between each occasion.
A circular wheel horn was attached to the servo motor axis,
which provided a rotational motion. A t-shaped rod was attached
to the wheel horn and this mechanical combination converted
the rotary motion of the motor to a linear motion of the rod (Fig.
4). The system provided a detectable indentation perpendicular
to the skin (5 mm indentation, 17 mm 2 area) with a force that
was sustained for 3 s. The distance between the stimulation
points on the skin was 40 mm; the minimal distance to detect
2-point discrimination on the forearm (29). The servo motors
were controlled by a microcontroller, Arduino Nano, which
acquired data from a graphical user interface developed in
LabVIEW (National Instruments, Austin, TX, USA) through
a serial interface.
The graphical user interface guides the subjects through 4
training sessions, which are described in detail in previous sec-
tion Learning. Prior to each session, the subjects got a descrip-
tive pop-up window about the coming session. The programme
was designed to be descriptive, to make sure that the subjects
could use it unsupervised. During the sessions, a picture of a
hand was shown. The subjects were instructed to select the
finger, using a mouse, onto which they associated the percei-
ved stimulation. Depending on which session was running, the
subject was given visual feedback about their performance (Fig.
2). The software logged the subjects’ information, such as age
and sex, along with each subject’s perceived stimulation value
and the actual stimulation value for each occasion and session.
At the end of every occasion the subject had the opportunity
to leave a comment about complications or other experiences
during the learning sessions.
Data analysis
In order to evaluate the agreement between the actual stimula-
tion and the response from the subjects, the linear weighted
Cohen’s kappa was calculated for each subject. By using a
linear weighted model, a response that is more distant finger-
wise to the actual stimulation were weighted more heavily than
a response that is closer to the actual stimulation. The strength
of the kappa value was assessed according to Brennan & Sil-
man (30); values < 0.20 are considered poor, values between
0.21–0.40 are considered fair, 0.41–0.60 moderate, 0.61–0.80
good and 0.81–1.00 very good. The kappa value was calculated
for each individual that participated and the median kappa value
and 95% confidence interval (95% CI) was calculated for each
training occasion.
To determine if the changes in the learning curve were sta-
tistically significant the Wilcoxon signed-rank test was used.
The kappa value was compared between paired observations;
occasion 1 (baseline), 11 (1 st day, week 2), 17 (2 nd follow-up,
week 4) and 18 (3 rd follow-up, week 5). These occasions were
chosen in order to analyse the learning progress between the
first and second week, and also if 1 week without training would
affect the new learned skill.
To assess if there were any differences between sexes the
2-tailed Mann–Whitney U test was used. This unpaired test
can determine the differences between 2 groups and it is also
useful in small groups (minimum 5).
To determine if there was a difference between age groups the
2-tailed Mann–Whitney U test was used. The age groups were
divided into 4 different groups; 20–29 (1 male and 10 females),
30–39 (2 males and 7 females), 40–49 (2 males and 2 females)
and > 50 years (1 male and 6 females).
Both pre-processing of data and analysis were performed in
Python, using packages such as Pandas (https://pandas.pydata.
org/) SciPy (https://www.scipy.org/) and scikit-learn (http://
scikit-learn.org/stable/).
For the analysis, the first evaluation session in every occasion
was chosen to evaluate the progression of learning, which shows
progression from the previous occasion rather than comparing
the progress within a single training occasion.
RESULTS
Fig. 4. Cross-section of the 3D-printed box, containing a servo motor
with a circular horn, which provides with a linear motion together with
the plastic rod, which in turn gives mechanotactile feedback on the skin.
www.medicaljournals.se/jrm
Of the 35 subjects, 31 completed the study (25 women
and 6 men). All but 2 were right-handed. The median
age was 37 (range 22–66) years. The 4 individuals who
did not complete the study dropped out at an early stage
without having to state a reason.
The training protocol was structured, but, in some
cases, there were minor discrepancies in the program-
me, as the subject performed the study unsupervised.
A few days of delays in the programme were recorded
(2–7 days) for some subjects. In total, 18 learning oc-
casions were planned in the programme, and among
the subjects, 16–19 learning occasions were recorded.
The cause of this was either due to technical problems,