ARTICLE
Auteur(s) : Johannes Gassmueller, Elisabeth Rowold,
Thomas Frase, Betsy
Hughes-Formella
bioskin GmbH, Burchardstraße 17, D-20095 Hamburg, Germany
accepté le 24 Decembre 2008
In 2001 TrichoScan® was introduced as a
fully-automated method for the measurement of biological parameters
of hair growth such as hair density or hair diameter [1]. The
method combines epiluminescence microscopy (ELM) with automatic
digital image analysis for the measurement of human hair [2].
Whereas in the past the dermatologist or technician literally had
to measure the length of the hair fibers and diameters with a
ruler, image analysis programs now allow for easy analysis of the
marked hairs. However, the manual identification of the hairs is a
tedious process prone to human error, even though manual
identification of hairs is sometimes defined as the most precise
method of measurement [3]. This might be the case when only the
number of hairs in a very small scalp area is counted, but in large
areas it is difficult for even well-trained technicians to make
accurate manual counts of the total, terminal, and vellus hairs as
well as hair thickness and hair growth rate. This must result in
variable results when the same image is counted two or more times.
Since at present no real independent side-by-side comparison is
available. Therefore, the aim of this study was the validation of
the TrichoScan® method by comparative assessment of hair
growth parameters using TrichoScan® software versus
manual identification of hairs prior to the final assessment of
hair parameters.
Materials and methods
Study participants
Digital images for TrichoScan® or conventional visual
analysis were taken from 10 patients aged 18 years or older with
AGA, Norwood-Hamilton grade III-IV/Ludwig grade 1 or 2. The
participants were selected from the volunteer panel at bioskin
GmbH, Hamburg. All patients included in this study had also taken
part in a previous hair growth study at bioskin and already had a
suitable measurement area marked with a tattoo. There were no other
pigmented lesions in the treatment area. The recommendations of the
Helsinki Declaration and the ICH GCP guidelines were followed.
Written informed consent was obtained before inclusion in the
study.
Tracking of target area and hair clipping
On day 1 the measurement area on the anterior border of the vertex
balding spot was identified. The area was clipped evenly (Moser,
TrichoScan Edition) and short clipped hair was removed by pressing
an adhesive strip onto the shaved area three times. The quality was
checked with a magnifying glass. Afterwards a digital image was
taken for documentation of the time. Digital images were stored in
an image database (Image DB).
Hair dye and TrichoScan® image on day
3
On day 3 (48 ± 2 hours after hair clipping) the clipped hairs
within the target area were dyed (Goldwell topchic, black 2N,
Darmstadt, Germany with Rondo 6% Crème-Oxyd, Coiffeur, Cologne,
Germany). After 12 minutes the colored area was thoroughly cleaned
with an alcoholic solution (Kodan® Spray, Schülke &
Mayr, Vienna, Austria) and digital images were taken using a
digital ELM system. Three separate images (B1, B2, B3) were taken
by 3 investigators (U1, U2, U3).
Evaluability criteria – Images for TrichoScan®
analysis
To be admissible the following requirements had to be fulfilled for
all images: All hairs were uniformly dyed, all hairs were evenly
clipped, no remnants of hair dye were present, no air bubbles were
present around the hairs, the image was bright and sharp, no hairs
from outside the measurement area crossed the field, and all hairs
were straight. None of the images which were taken had to be
excluded from analysis.
TrichoScan® analysis and generation
of data base
All images were analyzed using TrichoScan® Research
Edition 3.0 and results were imported into Excel. Data obtained by
manual evaluation were extracted with special software into a tab
delimited text file which was also imported into an
Excel® data sheet. The statistical analysis was
performed at bioskin.
Conventional image analysis by hand (manual
evaluation)
Three CDs of the images were produced for manual analysis by three
independent evaluators. These CDs contained the same images which
were analyzed by TrichoScan®, but they were additionally
embedded in a software program (“hair measure tool” provided by
Datinf GmbH, Tübingen). This software contained all 90 images in
random order. Randomization was done by Datinf GmbH, Tübingen,
Germany. The images were numbered 1, 2, (x), – 90. No
information was given about who took the image and from which
subject the image was taken. It was not possible to delete images
from or to add images to this CD. Each evaluator used a computer
mouse to outline the perimeters of each hair fiber on each image.
He/she had to click on every hair where the hair left the scalp
skin, then follow the hair and release the mouse button at the end
of the hair tip. All such marked hairs then appeared in yellow in
the software program (figure 1). The thickness
of the yellow line (the hair) was adjusted to the actual hair
thickness with the scroll wheel on the computer mouse. When the
yellow line had the same thickness as the actual underlying hair,
the correct thickness of this hair was determined. Hair density
(number per unit area on the image) and hair thickness (hair
diameter) were recorded automatically by the software. Hairs
starting from outside the target area which had the hair tips
inside the target area were not counted. Hairs which started inside
the target area but left it were counted, however these were not
used for the analysis of hair growth as the complete hair shaft was
not in the target area. The evaluator did not have access to any of
the calculated results such as hair density and thickness,
therefore he/she was unable to compare different analyses. After
each analysis the manual evaluator had to click “finish” and
thereafter this image was no longer available for counting. No
information was given about the results of the computerized
TrichoScan® images.
Data analysis and statistics
Study objectives
Validity
To prove the validity of TrichoScan® it was necessary to
show a strong correlation with manual evaluation. All analyses were
done with images as unit of observation and with patients as unit
of observation. The former ensured that all images were analyzed
without respect to multiple images of one subject. However, since
in practice not the images but the patients are the relevant
reference, it was also necessary to do the analysis for the values
averaged for patients.
Reliabilty
To evaluate the reliability, three different sources of variation
had to be taken into account: the patient, the investigator taking
the image, and the evaluator doing the manual evaluation or the
TrichoScan® software. Analysis of variability was done
separately for the manual evaluation of hair parameters and
TrichoScan®.
Number of evaluations
Images were taken from 10 patients. For each patient three
investigators took three images each on day 3, for a total of 90
different images. Each image was analyzed manually by three
different evaluators (270 manually analyzed images) as well as
three times with TrichoScan®. In addition, the first
image for each patient was made by Investigator 1 two more times.
This means that those images were analyzed manually three times.
The same was done with TrichoScan®. A total of 660
evaluations were performed, 50% by hand, 50% by
TrichoScan® (tables 1 and 2).
The mean hair density was 199/cm2 and
223/cm2, respectively. As we analyzed an area of
1.42 cm2, this adds up to 282 and 316 hairs/image.
This number multiplied by the number of evaluations amounts to a
total of 198,018 analyzed hairs.
Table 1 Overview of images, evaluation/analysis of
images and additional repetitions: Number of different images was
90 = 10 × 3 × 3 (product of no. of levels of SN, Inv and Img);
Number of evaluations without “Additional” was 90 × 2 × 3 = 540
(product of different images and no. of levels of Met and EvalNo);
Number of “Additional” evaluations was 10 × 2 × 3 × 2 = 120
(product of different images and no. of levels of Met, EvalNo and
repetitions (2))
|
Description
|
Variable
|
Levels
|
No. Levels
|
|
Images
|
Patient number
|
SN
|
1.10
|
10
|
|
Investigator defined as person who took the images
|
Inv
|
1,2,3
|
3
|
|
Repetition of image from Investigator
|
Img
|
1,2,3
|
3
|
|
Evaluation/Analysis of images
|
Method, TrichoScan® (TS) or manual analysis (Hand) of
image
|
Met
|
TS, Hand
|
2
|
|
No. of evaluation (three TrichoScan® analyses and three
different evaluators for manual hair counting
|
EvalNo
|
1,2,3
|
3
|
|
Additional
|
Repetition of evaluation – only for Met Hand, Img 1 and Inv 1, i.e.
two additional repeated analyses
|
RepNo
|
1,2,3
|
3
|
Table 2 Overview of data resulting from a) different
images and b) additional evaluation of image 1 by Investigator 1 to
investigate the effect of repeated analyses of identical images
|
a) All different images (90)
|
|
Met
|
Eval No.
|
Rep No.
|
No. of evaluations
|
|
TrichoScan
|
1
|
1
|
90
|
|
TrichoScan
|
2
|
1
|
90
|
|
TrichoScan
|
3
|
1
|
90
|
|
Hand
|
1
|
1
|
90
|
|
Hand
|
2
|
1
|
90
|
|
Hand
|
3
|
1
|
90
|
|
Sum
|
|
|
540
|
|
b) Additional repeated analysis of images of Inv 1, Img 1
(10)
|
|
Met
|
Eval No.
|
Rep No.
|
No. of evaluations
|
|
TrichoScan
|
1
|
2
|
10
|
|
TrichoScan
|
1
|
3
|
10
|
|
TrichoScan
|
2
|
2
|
10
|
|
TrichoScan
|
2
|
3
|
10
|
|
TrichoScan
|
3
|
2
|
10
|
|
TrichoScan
|
3
|
3
|
10
|
|
Hand
|
1
|
2
|
10
|
|
Hand
|
1
|
3
|
10
|
|
Hand
|
2
|
2
|
10
|
|
Hand
|
2
|
3
|
10
|
|
Hand
|
3
|
2
|
10
|
|
Hand
|
3
|
3
|
10
|
|
Sum
|
|
|
120
|
Analyzed variables
The following variables were evaluated manually and by TrichoScan®
for each image: TotalDens (total hair density, n/cm2);
TerminalDens (density of hairs thicker than 40 μm,
n/cm2); CumThickTotal (cumulative thickness of all
hairs, mm/cm2); CumThickTerm (cumulative thickness of
all terminal hairs, mm/cm2); MeanThickTotal (mean
thickness of all hairs, μm); MeanThickTerm (mean thickness of all
terminal hairs, μm); GrowthRateTotal (mean length of all hairs,
mm/day); GrowthRateTerm (mean length of all terminal hairs,
mm/day); CumGrowthRateTotal (sum length of all hairs, mm/day); and
CumGrowthRateTerm (sum length of all terminal hairs, mm/day).
Statistical methods
Validity
Descriptive statistics were performed, including differences
between the two methods of evaluation (mean and standard
deviation). In addition, a paired t-test (two-sided) was performed
for the differences, and correlation coefficients (Pearson) were
calculated. The analyses were done with images as the unit of
observation and with patients as the unit of observation (i.e. the
mean value calculated for each patient). Data from the repeated
evaluations of image 1 of Investigator 1 were averaged before
performing statistical tests.
Reliabilty
The analysis of the variation attributable to the investigator and
evaluator was done according to Bland and Altman [4]. Briefly, the
calculated variances were the variance between subjects
(σ2b), observers (σ2o),
different observers for different subjects
(σ2h) and variance of observations by one
observer for one subject (σ2w).
A two-way analysis of variance (ANOVA) with the factors
subject and observer and the interaction of subjects and observers
was calculated. The intra-observer variability was
σ2w, the inter-observer variability was
σ2o + σ2h +
σ2w. The repeatability by the same observer
was calculated as 2.83 × σw and the reproducibility when
different observers were used was calculated as 2.83 × square root
(σ2o + σ2h +
σ2w). Lower values reflect high repeatability
and reproducibility, higher values low repeatability and
reproducibility. The repeatability (and reproducibility) was an
estimate of the maximum difference (the limit within which 95% of
differences will lie) which can be obtained between two
measurements made at random on the same subject. The intra-class
correlation (ICC) coefficient for a single observer was calculated
as σ2b/(σ2b +
σ2w) and the ICC for different observers was
calculated as σ2b/σ2.
Intra- and inter-investigator variability and reliability. The
first image of Investigator 1 was evaluated three times by both
methods and the mean of the three repeated evaluations was
calculated. The analysis was carried out for each evaluation
method: Hand 1, Hand 2, Hand 3, TS 1, TS 2, TS 3.
Inter-evaluator variability. The first image of Investigator 1
was evaluated three times and the mean of the three repeated
evaluations was calculated. The mean from all three images taken by
the investigator for one subject was used in the ANOVA. Since data
from three different investigators were available, this analysis
was performed three times using the data from each investigator. In
addition the analyses were performed using the mean value of the
investigators.
Intra- and inter-evaluator variability and reliability.
A data subset was created using the first image from
Investigator 1 with the three repetitive evaluations by the three
evaluators. These data were used to calculate the intra- and
inter-evaluator variability and reliability.
Overall (investigator and evaluator) variability and
reliability. The overall variability and reliability were derived
from the investigator and evaluator variability and
reliability.
Variation coefficients. Variation coefficients were calculated
from the intra-evaluator variability and the corresponding mean
values. The results are given as percentages.
Results
Validity – Comparison of manual evaluation
and TrichoScan®
There was a very strong correlation between the manually evaluated
hair parameters and TrichoScan®, as evidenced by the
high Pearson correlation coefficients. All correlations with
patients as unit of observation were greater than 0.89 (0.85 with
images as unit of observation). All correlation coefficients were
highly significant (p < 0.001). The results are listed in table 3.
Table 3 Comparison of manually (Hand) and
TrichoScan® (TS) evaluated hair parameters: For each
parameter the mean, the difference, and the relative difference in
% are given for Hand and TS. These values are identical for the
analysis with images as unit of observation and with patients as
unit of observation. The correlation coefficient (Pearson) and the
p value for the paired t-test between the two methods are given for
the analysis with images as unit of observation and with patients
as unit of observation
|
Data
|
|
|
|
|
Images (N = 90)
|
Patients (N = 10)
|
|
Parameters
|
Mean TS
|
Mean Hand
|
Diff TS-Hand
|
Rel. Diff TS-Hand
|
Correlation
|
p (Diff)
|
Correlation
|
p (Diff)
|
|
Total hair density (n/cm2)
|
199.2
|
223.8
|
– 24.6
|
– 11.6%
|
0.967
|
< .0001
|
0.976
|
0.001
|
|
Density of hair thicker as 40 μm (n/cm2)
|
134.0
|
142.4
|
– 7.4
|
– 5.3%
|
0.956
|
< .0001
|
0.980
|
0.085
|
|
Cumulative thickness of all hairs (mm/cm2)
|
14.95
|
14.71
|
0.24
|
1.6%
|
0.982
|
0.003
|
0.996
|
0.057
|
|
Cumulative thickness of all terminal hairs (mm/cm2)
|
11.91
|
11.42
|
0.49
|
4.2%
|
0.943
|
0.002
|
0.971
|
0.200
|
|
Mean thickness of all hairs (μm)
|
54
|
48
|
6
|
12.3%
|
0.894
|
< .0001
|
0.940
|
< .0001
|
|
Mean thickness of all terminal hairs (μm)
|
62
|
56
|
6
|
10.5%
|
0.853
|
< .0001
|
0.894
|
< .0001
|
|
Mean length of all hairs divided by time difference from shaving
(mm/day)
|
0.448
|
0.444
|
0.003
|
0.8%
|
0.978
|
0.093
|
0.989
|
0.510
|
|
Mean length of all terminal hairs divided by time difference from
shaving (mm/day)
|
0.489
|
0.473
|
0.017
|
3.5%
|
0.980
|
< .0001
|
0.992
|
0.002
|
|
Sum length of all hairs divided by time difference from shaving
(mm/day)
|
120.1
|
129.6
|
– 9.5
|
– 7.6%
|
0.978
|
< .0001
|
0.983
|
0.008
|
|
Sum length of all terminal hairs divided by time difference from
shaving (mm/day)
|
90.19
|
89.87
|
0.32
|
0.4%
|
0.986
|
0.73
|
0.996
|
0.887
|
Reliability
Intra- and inter-investigator variability
and reliability
A general impression can be gained from the results illustrated in
figure 2 for the
plotted variables of total hair density and cumulative hair
thickness. The greatest variance was associated with the patients.
Therefore, the inter- and intra-investigator correlation
coefficients are all in a high range. Furthermore, the variability
for each patient was considerably higher for the individual manual
evaluators than for TrichoScan®, resulting in lower
repeatability and reproducibility, as well as lower intra-class
correlation coefficients (ICC) for a single and for different
investigators. The values are listed in table
4. The repeatability for the total hair density was
approximately three to five times higher for TrichoScan®
than for the three manual evaluators. Similar results were obtained
for the other variables. In table 4 the
results are listed and it can be seen that in general
TrichoScan® showed the highest intra-class correlation
coefficients and the greatest repeatability and reproducibility.
There was one exception: The “mean length of all hairs divided by
time difference from shaving (mm/day)” was similar for
TrichoScan® and the manual evaluators.
Table 4 Results of all parameters for the inter- and
intra-investigator variability for the three manual evaluations
Hand 1, Hand 2 and Hand 3 and the three TrichoScan®
evaluations which are identical and therefore listed only once as
TS 1, 2, 3. For each parameter and evaluation the repeatability
(maximum of the difference between two measurements on the same
patient by the same investigator), the reproducibility (maximum of
the difference between two measurements on the same patient by
different investigators) and the corresponding intra-class
correlation coefficients (ICC) are given
|
Parameter
|
Method
|
Repeatability by same investigator
|
Reproducibility by different investigators
|
ICC for a single investigator
|
ICC for different investigators
|
|
Total hair density (n/cm2)
|
Hand 1
|
33.2
|
33.2
|
0.969
|
0.969
|
|
Hand 2
|
39.3
|
40.5
|
0.956
|
0.953
|
|
Hand 3
|
61.0
|
61.0
|
0.897
|
0.897
|
|
TS 1, 2, 3
|
11.9
|
13.2
|
0.994
|
0.993
|
|
Density of hair thicker as 40 μm (n/cm2)
|
Hand 1
|
39.1
|
39.2
|
0.918
|
0.918
|
|
Hand 2
|
42.7
|
42.7
|
0.916
|
0.916
|
|
Hand 3
|
45.5
|
45.8
|
0.907
|
0.906
|
|
TS 1, 2, 3
|
15.7
|
18.8
|
0.981
|
0.972
|
|
Cumulative thickness of all hairs (mm/cm2)
|
Hand 1
|
2.77
|
2.77
|
0.940
|
0.940
|
|
Hand 2
|
2.95
|
3.00
|
0.944
|
0.943
|
|
Hand 3
|
3.64
|
3.68
|
0.919
|
0.917
|
|
TS 1, 2, 3
|
1.21
|
1.48
|
0.989
|
0.984
|
|
Cumulative thickness of all terminal hairs (mm/cm2)
|
Hand 1
|
3.16
|
3.16
|
0.926
|
0.925
|
|
Hand 2
|
4.36
|
4.36
|
0.889
|
0.889
|
|
Hand 3
|
4.18
|
4.27
|
0.897
|
0.894
|
|
TS 1, 2, 3
|
1.57
|
1.95
|
0.977
|
0.965
|
|
Mean thickness of all hairs (μm)
|
Hand 1
|
0.0060
|
0.0062
|
0.914
|
0.909
|
|
Hand 2
|
0.0106
|
0.0107
|
0.810
|
0.810
|
|
Hand 3
|
0.0080
|
0.0087
|
0.869
|
0.847
|
|
TS 1, 2, 3
|
0.0030
|
0.0040
|
0.968
|
0.944
|
|
Mean thickness of all terminal hairs (μm)
|
Hand 1
|
0.0050
|
0.0050
|
0.882
|
0.882
|
|
Hand 2
|
0.0077
|
0.0077
|
0.811
|
0.811
|
|
Hand 3
|
0.0045
|
0.0052
|
0.915
|
0.890
|
|
TS 1, 2, 3
|
0.0029
|
0.0038
|
0.962
|
0.935
|
|
Mean length of all terminal hairs divided by time difference from
shaving (mm/day)
|
Hand 1
|
0.0415
|
0.0459
|
0.975
|
0.969
|
|
Hand 2
|
0.0412
|
0.0426
|
0.978
|
0.976
|
|
Hand 3
|
0.0529
|
0.0544
|
0.962
|
0.960
|
|
TS 1, 2, 3
|
0.0309
|
0.0332
|
0.985
|
0.983
|
|
Mean length of all hairs divided by time difference from shaving
(mm/day)
|
Hand 1
|
0.0219
|
0.0258
|
0.992
|
0.990
|
|
Hand 2
|
0.0305
|
0.0351
|
0.986
|
0.982
|
|
Hand 3
|
0.0439
|
0.0467
|
0.971
|
0.968
|
|
TS 1, 2, 3
|
0.0287
|
0.0318
|
0.984
|
0.981
|
|
Sum length of all hairs divided by time difference from shaving
(mm/day)
|
Hand 1
|
16.2
|
16.2
|
0.985
|
0.985
|
|
Hand 2
|
26.4
|
27.0
|
0.963
|
0.961
|
|
Hand 3
|
40.5
|
40.5
|
0.914
|
0.914
|
|
TS 1, 2, 3
|
8.5
|
9.5
|
0.996
|
0.995
|
|
Sum length of all terminal hairs divided by time difference from
shaving (mm/day)
|
Hand 1
|
18.6
|
18.7
|
0.980
|
0.979
|
|
Hand 2
|
24.8
|
24.9
|
0.967
|
0.967
|
|
Hand 3
|
26.2
|
26.2
|
0.962
|
0.962
|
|
TS 1, 2, 3
|
9.7
|
12.1
|
0.993
|
0.989
|
Intra- and inter-evaluator variability
and reliability
Values for inter- and intra-evaluator variability can be found in
table 5. As expected for a
fully-automated system, there were no differences for
TrichoScan® on repeated evaluations. Therefore the
repeatability and reproducibility were always equal to 0 and the
corresponding intra-class correlation coefficient was always equal
to 1. This was not the case for the manual evaluation. In addition,
the variation coefficients for intra-evaluator variability were
calculated. Mean data variability in hand evaluated images for
terminal hair thickness ranged up to 12.95%, whereas the
TrichoScan® variability was zero (table 6).
Table 5 Results of all parameters for the inter- and
intra-evaluator variability for the manual evaluation (Hand) and
the TrichoScan® (TS) evaluation. For each parameter and
method the repeatability (maximum of the difference between two
measurements on the same patient by the same evaluator), the
reproducibility (maximum of the difference between two measurements
on the same patient by different evaluators) and the corresponding
intra-class correlation coefficients (ICC) are given
|
Parameter
|
Method
|
Repeatability by same evaluator
|
Reproducibility by different evaluators
|
ICC for a single evaluator
|
ICC for different evaluators
|
|
Total hair density (n/cm2)
|
|
|
|
|
|
|
Density of hair thicker as 40 μm (n/cm2)
|
|
|
|
|
|
|
Cumulative thickness of all hairs (mm/cm2)
|
|
|
|
|
|
|
Cumulative thickness of all terminal hairs (mm/cm2)
|
|
|
|
|
|
|
Mean thickness of all hairs (μm)
|
|
|
|
|
|
|
Mean thickness of all terminal hairs (μm)
|
|
|
|
|
|
|
Mean length of all terminal hairs divided by time difference from
shaving (mm/day)
|
|
|
|
|
|
|
Mean length of all hairs divided by time difference from shaving
(mm/day)
|
|
|
|
|
|
|
Sum length of all hairs divided by time difference from shaving
(mm/day)
|
|
|
|
|
|
|
Sum length of all terminal hairs divided by time difference from
shaving (mm/day)
|
|
|
|
|
|
Table 6 Data variability of manual evaluation (Hand)
and TrichoScan®. The data variability of one
investigator is shown
|
Variable
|
Hand
|
TrichoScan®
|
|
Hair Count Total
|
7.07%
|
0.00%
|
|
Hair Count Terminal
|
11.41%
|
0.00%
|
|
Cumulative Thickness Total
|
8.09%
|
0.00%
|
|
Cumulative Thickness Terminal
|
12.95%
|
0.00%
|
|
Mean Thickness
|
6.53%
|
0.00%
|
|
Mean Thickness Terminal
|
3.99%
|
0.00%
|
|
Growth Rate Total
|
2.71%
|
0.00%
|
|
Growth Rate Terminal
|
3.38%
|
0.00%
|
|
Cumulative Growth Rate Total
|
8.12%
|
0.00%
|
|
Cumulative Growth Rate Terminal
|
9.43%
|
0.00%
|
Discussion
The present study was designed to compare the variability of a
semi-automated procedure with manual identification of hairs prior
to analysis by hair growth software with a fully-automated
procedure with recognition of hairs by the software. In this
GCP-conform study the clinical trial situation was imitated where
different investigators use the same equipment on different
patients.
In this study there was a highly significant correlation between
evaluation of hair parameters using manual identification of hairs
and the fully-automated TrichoScan® method. This
demonstrates that the TrichoScan® software, although
working by statistics and mathematical approximation, counts hairs
and not artefacts. Nevertheless, there are some differences between
the methods concerning absolute values. The strongest differences
were seen in the mean thickness and total density of the hairs.
TrichoScan® values for hair thickness were approximately
10% higher and density values approximately 10% lower than the
values obtained by the manual evaluators. The differences in most
other parameters were less than 5%. In particular, the critical
parameter cumulative hair thickness was in good agreement with the
manual evaluation. It is not surprising that TrichoScan®
underestimates the total hair density: As a digital tool
TrichoScan® relies on camera resolution. In this study
resolution was 2 Megapixel, a resolution at which very thin (below
7 μm) hairs are not analyzed. In our opinion this is not a
disadvantage as these hairs are not clinically relevant. Regarding
the overestimation of hair thickness by TrichoScan®, it
must be kept in mind that this is a systematic difference and
occurs for every hair. Therefore, it has no impact on the
TrichoScan® capability to measure changes in hair
thickness. If needed, the thickness can be mathematically
readjusted.
Not surprisingly, considerable variability was noted for the
manually marked images. Manual marking of hairs is tedious and
time-consuming, making it near to impossible to repeatedly count
hundreds of hairs without variation for density, length and
thickness. The mean data variability for one evaluator who marked
the same image three times ranged from 2.71-12.95%, depending on
the parameter. Some evaluators showed more variability than others.
The correlation between different evaluators was best for total
hair density and parameters related to hair length and worst for
the parameters related to hair thickness. There was no variability
in repeated measures with TrichoScan®, the software
delivered completely reliable results. Keeping in mind the
definition of repeatability, the maximum of the difference between
two measurements on the same patient, this is a tremendous
difference between manual marking of hairs and
TrichoScan®. In a clinical trial setting this would mean
that a much larger sample size is required in the case of manual
marking compared to fully-automated evaluation.
In order to analyze data reproducibility due to investigator
variability, the situation was compared when one investigator
captured a phototrichogram image of the same target area three
times or three different investigators captured the same target
area. The statistical data correspond to the intra- and
inter-investigator error related to the taking of images. Results
for all parameters measured using TrichoScan® software
verified that images made by one investigator were highly reliable,
with a very high correlation (ICC ≥ 0.962) for all parameters.
Likewise, images of the same target area made by different
investigators showed similar reproducibility. Although the manual
evaluations also produced very robust results, the data variability
introduced by manual measurements led to a lower overall data
reproducibility.
It must be kept in mind that the quality of the results with
TrichoScan® or any other imaging tool depends on the
quality of the images. Only technically correct images can deliver
good results. Although in our experience the TrichoScan®
procedure is relatively easy, some investigators may have dye
remnants, unfocused images, or large air bubbles in the image. In
clinical trials this may be overcome by adequate investigator
training and strict quality control of incoming images. To achieve
the quality of images used in this study approximately 3 hours of
training were required per investigator or study nurse.
In summary, in this validation study we saw an excellent
correlation of hair growth parameters analyzed using the
fully-automated TrichoScan® method and manual marking of
hairs prior to analysis. Considerable variability was seen in the
results from manually identified hairs, compared to none in
TrichoScan®-analyzed images. In a clinical trial, the
wider margin of error and consequent data variability from manually
evaluated images would necessitate a larger study sample size to
overcome the effect of the variability on the statistical
calculations. Therefore, the TrichoScan® technique is
particularly suitable for clinical studies with treatment
comparisons. That this is indeed the case can be seen in the
results of a recent clinical trial evaluating changes of hair
growth and thickness in 34 Minoxidil-treated men. Using the
TrichoScan® method, treatment benefits could be seen 8
weeks after treatment initiation [5]. Moreover,
TrichoScan® can be adopted to study the effect of drugs
or laser treatment on hypertrichosis or hirsutism [6]. This,
however, requires a different software algorithm which was outside
the scope of this trial.
Last but not least, it should be kept in mind that even though
phototrichograms or other hair analysis methods are important tools
in the evaluation of hair loss treatments, in later phase clinical
trials it is also important to assess other measures such as
quality of life.
Acknowledgements
bioskin GmbH as an independent CRO was contracted by the
TrichoScan® manufacturer (Tricholog GmbH, Freiburg,
Germany) to perform this validation trial. TrichoScan®
raw data were provided by Datinf GmbH, Tübingen, Germany and raw
data from the manual hair counting by Tricholog.
References
1 Hoffmann R. TrichoScan: combining epiluminescence microscopy
with digital image analysis for the measurement of hair growth in
vivo. Eur J Dermatol 2001; 11: 362-8.
2 Hoffmann R. TrichoScan: a novel tool for the analysis of
hair growth in vivo. J Investig Dermatol Symp Proc 2003; 8:
109-15.
3 Van Neste D, Trüeb RM. Critical study of hair growth
analysis with computer-assisted methods. JEADV 2006; 20:
578-83.
4 Bland JM, Altman DG. Measurement error. BMJ 1996;
313: 744.
5 Gassmueller J, Hoffmann R, Webster A. Treatment
of androgenetic alopecia using topical fulvestrant solution in
males and post-menopausal females; results from two randomized,
proof-of-concept studies. British J Dermatol 2008; 158: 109-15.
6 Hoffmann R. A 4-month, open-label study evaluating the
efficacy of eflornithine 11.5% cream in the treatment of unwanted
facial hair in women using TrichoScan. Eur J Dermatol 2008; 18:
65-70.
|
Version imprimable
Version PDF
