SKIN 
 

March 2019     Volume 3  Issue 2 
 

Copyright 2018 The National Society for Cutaneous Medicine 120 

RISING DERM STARS® 
 

 
Pilot study of use of neural network for quantitative before-and-after analysis 
of dermatologic cosmetic procedures 
 
Adam Aronson, MD1, Anastasia Georgievskaya, Hillary Jahangir-Johnson, MD, Ksenia Guseva 
1Department of Dermatology, University of Iowa, Iowa City, IA 

 
 
Background/Objectives: In dermatologic 
cosmetic procedures, quantitative measures 
of efficacy have historically been lacking. 
Before and after pictures provide an 
illustration of the effects of a procedure, but 
are often cherrypicked to show best 
outcomes rather than typical outcomes and 
are inherently subjective. Attempts to apply 
statistic analysis regarding degree of 
cosmetic improvement typically relies on 
various assessment scales, in which various 
metrics are subjectively graded on a Likert 
scale by two or more blinded scorers, and 
numerable results are then reconciled and 
statistical analysis is performed1,2,3. These 
methods are time-consuming and may 
require multiple revisions to attain acceptable 
inter- and intrarater reliability, which 
decreases study objectivity. 
 
Methods: Here, we present an ongoing pilot 
study exploring potential for deep learning 
neural networks to perform visual facial 
analysis both before and after dermatologic 
cosmetic procedures, with the goal of 
creating a more accurate and reproduceable 
understanding of which facial characteristics 
are affected by the procedure. Youth 
Laboratories (ylabs.ai) has developed and 
trained a neural network for evaluation of skin 
parameters with the goal of analyzing skin 
health and disease biomarkers. An earlier 
version of this software was publicly available 
as a downloadable application called 

RYNKL. This software has since been 
updated to tailor it to this project. It currently 
tracks parameters including skin evenness, 
redness, sebum, porphyrins, skintone, pore 
characteristics, and estimated age (Figure 1). 
 
In theory, this technology should be useful for 
measuring efficacy of almost any facial 
cosmetic procedure after appropriate 
validation. In the pilot study, we are focusing 
on full facial resurfacing procedures with 
CO2RE laser. Patients electively choosing to 
undergo this procedure are identified pre-
procedure and consent is obtained for 
photography and software analysis of 
images. At this time, enrollment is limited to 
Caucasian females, as the neural network 
currently has the most reference images for 
this demographic. High resolution facial 
photographs are obtained prior to procedure 
and at 3 months post-procedure. All of the 
above-listed parameters are numerically 
rated by the RYNKL software, with the goal 
of determining which quantitative differences, 
if any, are measurable after healing from the 
procedure. Following collection of all before 
and after photographs, four blinded 
dermatology residents will also score each 
photograph for the same parameters for 
comparison of reliability between trained 
scorers and neural network. Photograph and 
data collection is ongoing at the time of this 
writing. 
 



SKIN 
 

March 2019     Volume 3  Issue 2 
 

Copyright 2018 The National Society for Cutaneous Medicine 121 

Discussion: Though neural networks have 
not been trained for this purpose previously, 
they are not entirely new to dermatology. 
Neural networks have already been trained to 
be comparable to board-certified 
dermatologists in the identification of skin 
cancers4. Though this study is still in the proof 
of concept phase, neural networks by their 
nature improve incrementally with each new 
image fed to them and thus become more 
powerful every day. The use of neural 
networks for improving the depth of 
dermatologic research is one of many 
developing usages. 
 
Figure 1: 

 
 

References: 
1. Idriss N and Maibach HI. Scar 

assessment scales: a dermatologic 
overview. Skin Res Technol. 2009 
Feb;15(1):1-5. PMID: 19152571 

2. Carruthers A and Carruthers J. A 
validated facial grading scale: the future 
of facial ageing measurement tools? J 

Cosmet Laser Ther. 2010 
Oct;12(5):235-41. PMID: 20825260 

3. Narins RS et al. Validated assessment 
scales for the lower face. Dermatol Surg. 
2012 Feb;38(2 Spec No.):333-42. PMID: 
22316189 

4. Esteva A et al. Dermatologist-level 
classification of skin cancer with deep 
neural networks. Nature. 
2017;542(7639):115-118. PMID: 
28117445