Accepted Article

MISS JULIE ROBIC (Orcid ID : 0000-0002-0666-8728)

Article type : Letter to Editor

Clinical validation of a computer-based approach for the

quantification of the skin aging process of women using
in-vivo confocal microscopy

J. Robic1§, A. Nkengne1, B. Perret2, M. Couprie2, H. Talbot3, G. Pellacani4, K. Vie1

1 Laboratoires Clarins, 5 rue Ampère, 95300 Pontoise, France

2 Université Paris-Est, Laboratoire d'Informatique Gaspard-Monge UMR 8049, UPEMLV,
ESIEE Paris, ENPC, CNRS, F-93162 Noisy-le-Grand France
3 Université Paris-Saclay, CentraleSupélec, Inria, Centre de Vision Numérique, F-91190,
Gif-sur-Yvette, France
4 Department of Dermatology, University of Modena and Reggio Emilia, Italy
§ Corresponding author

Correspondance : Julie Robic

Laboratoires Clarins, 5 rue Ampère, 95300 Pontoise, France
Email: julie.robic@clarins.com
Phone: +33 (0)1 34 35 15 15

Keywords:
In-vivo confocal microscopy, skin aging, automated quantification, validation

This article has been accepted for publication and undergone full peer review but has not been
through the copyediting, typesetting, pagination and proofreading process, which may lead to
differences between this version and the Version of Record. Please cite this article as doi:
10.1111/JDV.16810
This article is protected by copyright. All rights reserved
 Word count: 570
Accepted Article
Table count: 1
Figure count: 1
References: 7

Funding
This work was supported by Clarins laboratories.

Competing interests
No conflicts of interest, financial or otherwise, are declared by the authors.

This article is protected by copyright. All rights reserved

 Dear Editor,
Accepted Article
Reflectance confocal microscopy (RCM) is a powerful tool to visualize the skin layers at
cellular resolution up to a depth of 200µm. A semi-quantitative score of skin aging from
RCM images has been previously published, requiring visual assessment of the images
by experienced dermatologists 1, 2. We have proposed in a previous publication, new
computer-based methods to automatically quantify the skin aging process on RCM
images 3-7. This letter presents the results of a clinical study designed to validate the
computer-based methods versus expert assessment of RCM skin images. The study
was conducted according to Good Clinical Practices and followed the Helsinki
convention .The protocol was approved by the internal ethics comity of the Modena and
Reggio Emilia Hospital, Italy. We enrolled 160 healthy women from 4 different ethnic
groups (Caucasian, Hispanic, African and Asian). RCM images were acquired on the
cheek using a near-infrared reflectance confocal laser scanning microscope. Clinical
annotations were evaluated by an expert dermatologist to assess the following aging
signs as described in 1, 2: epidermal honeycomb pattern; shape of the dermal papillae;
organization of the dermal fibers (Fig1).
The epidermal honeycomb pattern was annotated as either regular or irregular.
The proposed algorithm calculates the percentage of regular cells and the average size
of regular regions to provide scores at the epidermal layer 4.
The shape of the dermal papillae was clinically annotated as ringed, poly-cyclic or
both. This shape was computationally characterized by local and topological descriptors,
through their number, compactness and elongation 5-7. The last two tend to 1 for a
perfect circle. Sixty seven out of 160 subjects were used for the validation of the DEJ
algorithm, as they presented only one visible type of pattern, i.e. ringed or poly-cyclic.
Most subjects with visible dermal papillae had a dark photo-type.

Finally, collagen fibers type were manually annotated as being reticulated or

coarse. The machine learning approach developed 3 calculates the percentage of
reticulated fibers in an image and classify the images as reticulated or coarse.

For each skin layer, experts’ clinical annotations defined two populations. Means
of distributions of the algorithm parameters were compared between them with Student’s

This article is protected by copyright. All rights reserved

 or Mann-Whitney tests (p <0.05), depending on the result of the normal distribution test
Accepted Article
(Shapiro normality test). Random Forest classifiers were then used to create a predictive
model for clinical annotations from the computer-based descriptors. The accuracy,
sensitivity and specificity of these models were computed.

Regarding the epidermal layers, no statistical difference is found in the percentage

of regular cells between the two populations, i.e. honeycomb patterns annotated as
regular and irregular. On the contrary, the average size of regular regions is significantly
higher among regular honeycomb patterns. The clinical annotation prediction achieves an
average precision score of 80% while the sensitivity and specificity scores are slightly
above 80%, see Table 1.
The number, mean compactness and elongation of dermal papillae are
significantly higher among ringed dermal papillae than among poly-cyclic dermal papillae.
We achieve 83%, 76% and 81% for respectively the global accuracy, sensitivity and
specificity (Table 1).
As for the collagen fibers, the percentage of reticulated fibers among is
significantly higher in the reticulated population. We achieve a specificity of 89%.
Computer classification of the main skin parameters, as observed in RCM, showed
good correlation with the expert evaluator. To our knowledge, these are the first results
comparing a computer-based approach to dermatologists’ approach for the assessment
of skin aging using in vivo confocal microscopy.
References
1. Longo C, Casari A, Beretti F, Cesinaro AM, Pellacani G. Skin aging: in vivo
microscopic assessment of epidermal and dermal changes by means of confocal
microscopy. Journal of the American Academy of Dermatology. 2013;68(3):e73-e82.

2. Longo C, Casari A, De Pace B, Simonazzi S, Mazzaglia G, Pellacani G. Proposal

for an in vivo histopathologic scoring system for skin aging by means of confocal
microscopy. Skin Research and Technology. 2013;19(1):e167-e73.

3. Robic J. Automated characterization of skin aging using in vivo confocal

microscopy: Paris Est; 2018.

This article is protected by copyright. All rights reserved

 4.
Accepted Article Robic J, Nkengne A, Perret B, Couprie M, Talbot H, editors. Automated
quantification of the epidermal aging process using in-vivo confocal microscopy. 2016
IEEE 13th International Symposium on Biomedical Imaging (ISBI); 2016: IEEE.

5. Robic J, Perret B, Nkengne A, Couprie M, Talbot H, editors. Classification of the

dermal-epidermal junction using in-vivo confocal microscopy. 2017 IEEE 14th
International Symposium on Biomedical Imaging (ISBI 2017); 2017: IEEE.

6. Robic J, Perret B, Nkengne A, Couprie M, Talbot H. Three-dimensional conditional

random field for the dermal–epidermal junction segmentation. Journal of Medical
Imaging. 2019;6(2):024003.

7. Robic J, Perret B, Nkengne A, Couprie M, Talbot H, editors. Self-dual Pattern

Spectra for Characterising the Dermal-Epidermal Junction in 3D Reflectance Confocal
Microscopy Imaging. International Symposium on Mathematical Morphology and Its
Applications to Signal and Image Processing; 2019: Springer.

Legends of figures

Figure 1. Each image corresponds to a horizontal section with a 500 × 500µm field of
view. First row: Epidermal aging. (a) The epidermis exhibits a regular honeycomb pattern
(b) Irregular honeycomb pattern. Second row: DEJ aging. (c) Ringed dermal papillae (d)
poly-cyclic dermal papillae. Third row: Collagen fibers types. (e) Reticulated fibers
characterized by tiny reflective fibers orderly disposed forming web like structures, (f)
coarse fibers composed by thick fibers grossly arranged into nets.

Table legend
Table 1. Scores of the clinical annotations prediction for each skin layer.

This article is protected by copyright. All rights reserved

Accepted Article Mean of distributions
Clinical annotation prediction scores
Algorithm
Skin Layer for each population, i.e.
output
clinical annotation
Accuracy Sensitivity Specificity

Percentage of Regular / Irregular

regular cells 0,36 ± 0,08 / 0,32 ± 0,07

Average size of
Epidermis 80 81 81
regular regions
Regular / Irregular
(mean number
6.6 ± 3.3 / 5.2 ± 3.1
of neighboring
regular cells)

Number of Cyclic / Poly-cyclic

dermal papillae 24.9 ± 4.9 / 22 ± 4.7

Mean
compactness of Cyclic / Poly-cyclic
DEJ 83 76 81
dermal papillae 0.45 ± 0.01 / 0.43 ± 0.02
[0,1]
Mean elongation
Cyclic / Poly-cyclic
of dermal
0.58 ± 0.02 / 0.56 ± 0.02
papillae [0,1]

Collagen fibers
type Reticulated / Coarse
Dermis 80 63 89
(reticulated/coar 0,66 ± 0,22 / 0,38 ± 0,17
se)

Table 1. Scores of the clinical annotations prediction for each skin layer.

This article is protected by copyright. All rights reserved

Accepted Article

jdv_16810_f1.tif

This article is protected by copyright. All rights reserved