105 / 162
103
5
while the mean r was reduced at both ends of the array (to r = 0.86). R-values for electrodes 2, 5, 9 and 14
are not shown in figure 5b, since these electrodes were active in only less than 33% of the subjects (table 3).
The goodness of fit of equation 2 for the individual T-levels was tested in the remaining 30% of the measured
data (fig. 5c; table 3a). Figure 5c shows scatter plots of the predicted T-levels versus the measured T-levels
for all 16 electrode contacts, while table 3a provides the associated numerical data. Again, it is clear that the
predictions are slightly better for the center region of the array. This procedure was repeated with a number
of other random selections of 30% of the population, with essentially the same result.
To obtain a T-level profile expressed in clinical units, equation 2 can be reformulated as follows:
A fit comparable to figure 5c was made for the M-level profile (not shown), and, again, a high predictability
could be obtained with a measurement on only one electrode contact (table 3b). On the basis of a similar
mixed linear model, the M-levels along the array could be predicted with equations 4 and 5 (in decibels and
clinical units, respectively):
M-level profile (with emphasis on higher frequencies) was set during fitting in our clinic (see Subjects and
Methods).
The bars in figure 6a, b show the mean differences between the predicted and measured T-levels, while
the dashed lines indicate the lower and upper borders of the 95% prediction interval for the individual
electrode con-tacts, expressed in decibels (fig. 6a) and clinical units (fig. 6b). Figure 6c, d shows the same
data for the M-lev-els. It is clear that the size of the 95% prediction interval increases with the distance from
electrode contact 7, at which Tand M-levels are measured, in spite of the fact that the model predicts the
mean levels accurately alongthe whole array.
van der Beek/Briaire/Frijns
:1–16
de 7, yielding the
+
B
).
(2)
electrodes in the
e results, while the
array (to r = 0.86).
are not shown in
active in only less
the individual T-
of the measured
scatter plots of the
predicted T-levels versus the measured T-levels for all 16
electrode contacts, while table 3a provides the associated
numerical data. Again, it is clear that the predictions are
slightly be ter for the c nte region of the array. This pro-
cedure was repeated with a number of other random se-
lections of 30% of the population, with essentially the
same result.
To obtain a T-level profile expressed in clinical units,
equation 2 can be reformulated as follows:
2 2
2 2
1 0 01
7 0 025
20
7
0 2
7 0 5
7
7
-
10
10
.
lectrode
.
lectrode
lectrode
.
lectrode
.
lectrode
lectrode
T level lectrode
T-level
T-level
7
×
×
e
e
e
e
e
e
e
(in CU).
(3)
f the measured T-levels vs. predicted T-levels for the remaining 30% of the population for
Downloaded by:
Leiden University
145.88.209.33 - 11/23/2014 2:34:38 PM
van der B
Audiol Neur tol 2015;20:1–16
DOI: 10.1159/000362779
12
A fit comparable to figure 5c was made for the M-level
profile (not shown), and, again, a high predictability
could be obtained with a measurement on only one elec-
trode contact (table 3b). On the basis of a similar mixed
linear model, the M-levels along the array could be pre-
dicted with equations 4 and 5 (in decibels and clinical
units, respectively):
(4)
7
1 8 cos 15
cos 2
1
32
32
0 118
7
and
lectrode
M-level lectrode .
lectrode
.
lectrode
M-level
¬
¬
¬
®
®
®
e
e
e
e
(
in dB
)
(5)
1 1 8 cos 15
cos 2
1
0 118
7
20
32
32
7
36 cos 15
cos 2
1
32
32
7
10
10
.
lectrode
.
lectrode
lectrode
lectrode
lectrode
M-level lectrode
M-level
M-level
¬
¬
¬
¬
®
®
®
®
¬
®
2 36
7
.
lectrode
¬
¬¬
®
®
®
e
×
×
e
e
e
e
e
e
(in CU).
Contrary to the situation for the T-levels, which were
measured individually, this high predictability for the M-
level profile was mainly due to the fact that a predefined
M-level profile (with emp
was set during fitting in our
ods).
The bars in figure 6a, b s
tween the predicted and
dashed lines indicate the lo
95% prediction interval for
tacts, expressed in decibel
(fig. 6b). Figure 6c, d shows
els. It is clear that the size o
increases with the distance
which T- and M-levels are
that the model predicts the
the whole array.
Discussion
The present paper demo
of T- and M-levels in a relat
described in closed-set for
as a starting point for fittin
T-level
–6
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
T-level
–60
–40
a
c
b
M-level – predicted M-level (dB)
–6
–3
0
3
6
M-level – predicted M-level (CU)
–60
–40
–20
0
1 2 3 4 5 6 7
20
40
60
d
Fig. 6.
Prediction error means (bars) and 95% prediction intervals (dashed lin
clinical units (
b
) as well as of M-levels in decibels (
c
) and clinical units (
d
).