Validation of a new facial skin analysis device across Fitzpatrick skin types

AI skincare is a skin analysis system validated with the Canfield VISIA®. 1,2 Self-learning algorithms may underperform when analyzing skin of color, as most training material is collected from fair-skinned populations. 3,4 The objective of this study was to assess the validity of the skin analysis system across Fitzpatrick skin types. Adult participants were recruited during a routine dermatology appointment. After obtaining informed consent, facial analyses with AI Skincare (Perfect Corporation) and VISIA® Complexion Analysis model Generation 7 (Canfield Scientific, Inc.) were performed. AI skincare is an iOS tablet application that takes a frontal facial photo-graph. VISIA® is a computer-based system that utilizes a facial imaging booth. Both instruments provide percentiles for spots, wrinkles, texture, pores, and redness. Lower scores indicate less desirable characteristics. Fitzpatrick skin types were obtained and Types I and II, III and IV, and V and VI were categorized into Group 1, Group 2, and Group 3, respectively. For each metric, the relationship between AI skincare and VISIA® scores for each Fitzpatrick group was assessed with a Pearson correlation coefficient. Fisher r to z transformations were conducted to assess for equality between correlation coefficients. The tablet and computer-based assessments were compared with a physician's assessment. Photographs of participants were numbered and paired consecutively. The participant in each pair with poorer health was recorded for each metric. If a pair had the same score for a particular metric, it was removed from the analysis for that metric. A board-certified dermatologist (S.R.F.) blinded to the facial analysis scores reviewed the paired photographs and recorded which participant had worse skin health for each metric. These rankings were then compared to the facial analysis systems' rankings to determine concordance.


L E T T E R S T O T H E E D I T O R Validation of a new facial skin analysis device across Fitzpatrick skin types
AI skincare is a skin analysis system validated with the Canfield VISIA®. 1,2Self-learning algorithms may underperform when analyzing skin of color, as most training material is collected from fairskinned populations. 3,4The objective of this study was to assess the validity of the skin analysis system across Fitzpatrick skin types.
Adult participants were recruited during a routine dermatology appointment.After obtaining informed consent, facial analyses with AI Skincare (Perfect Corporation) and VISIA® Complexion Analysis model Generation 7 (Canfield Scientific, Inc.) were performed.AI skincare is an iOS tablet application that takes a frontal facial photograph.VISIA® is a computer-based system that utilizes a facial imaging booth.Both instruments provide percentiles for spots, wrinkles, texture, pores, and redness.Lower scores indicate less desirable characteristics.Fitzpatrick skin types were obtained and Types I and II, III and IV, and V and VI were categorized into Group 1, Group 2, and Group 3, respectively.
For each metric, the relationship between AI skincare and VISIA® scores for each Fitzpatrick group was assessed with a Pearson correlation coefficient.Fisher r to z transformations were conducted to assess for equality between correlation coefficients.
The tablet and computer-based assessments were compared with a physician's assessment.Photographs of participants were numbered and paired consecutively.The participant in each pair with poorer health was recorded for each metric.If a pair had the same score for a particular metric, it was removed from the analysis for that metric.A board-certified dermatologist (S.R.F.) blinded to the facial analysis scores reviewed the paired photographs and recorded which participant had worse skin health for each metric.
These rankings were then compared to the facial analysis systems' rankings to determine concordance.
Fifty-one participant pairs were evaluated across five metrics (255 total points of comparison).After duplicate scores were removed, 240 and 252 points of comparison remained in the tablet and computer-based application groups, respectively.The tablet and computer applications' rankings were the same as the physician's ranking 69% and 56% of the time, respectively.When rankings were stratified by Fitzpatrick group, the tablet application had superior agreement with the physician's assessment for all metrics, except for pores in Groups 1 and 2 (Table 2).
The tablet application produced similar results to the standard computer-based application and may be superior for analyzing skin of color, as it outperformed the latter when compared with the physician's assessment.Thus, the tablet application is a valid instrument for skin analysis across patients with all skin types.Limitations include a small sample size and only including a single physician assessment.
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.© 2023 The Authors.Journal of Cosmetic Dermatology published by Wiley Periodicals LLC.Equality of Pearson's correlation coefficients based on Fitzpatrick skin types.
TA B L E 1