TITLE:
CTK Derma AI: Any Skin Tone-Aware AI Capture Calibration for Objective Quantification of Facial Aging Features: Validation across Fitzpatrick Phototypes I - VI
AUTHORS:
Shu Li, Labina Shrestha, Mirjalol Tuychiev, Tae Yong Jung, Ryan Wonsuk Choi, Kukizo Miyamoto
KEYWORDS:
Skin Diagnosis, Skin Aging, Wrinkle, Pore, Spots, CTK Derma AI, Skin Phototype
JOURNAL NAME:
Journal of Cosmetics, Dermatological Sciences and Applications,
Vol.16 No.3,
August
3,
2026
ABSTRACT: Background: Image-based assessment of facial aging signs is sensitive to lighting, device-dependent color rendering, and skin phototype, which can compromise reproducibility and fairness across diverse populations. Objective: To validate a skin tone-aware AI-based image capture calibration and analysis framework (CTK Derma AI) for the objective quantification of pores, pigmented spots, and wrinkles across Fitzpatrick phototypes I - VI, using expert visual grading as the reference standard. Methods: Facial images from 265 participants were captured using a handheld CTK ChoiceDx® platform device under a standardized imaging protocol. Participants were grouped into Fitzpatrick I - II (n = 58), III - IV (n = 61), and V - VI (n = 146). Regions of interest included the periorbital area for wrinkle analysis and the central cheek for pore, spot and skin tone (L-value) analysis by the image analysis, and then graded by the expert graders. CTK Derma AI analysis framework was trained by half of the dataset (n = 132), then validated using the remaining dataset (n = 133). CTK Derma AI pipeline was compared with a conventional non-calibrated image-processing baseline (called Control, hereafter). Scores were normalized to a 0 - 100 scale, with higher scores indicating fewer visible aging signs. Correlations with expert grading were assessed using Pearson’s r; differences between dependent correlations were pre-specified for R-based Steiger/ Meng-style testing. Repeatability was evaluated across three repeated captures per subject. Results: CTK Derma AI showed statistically stronger overall correlations with expert grading than Control for pores, spots, wrinkles and skin tone. The largest improvements were observed in the Fitzpatrick V - VI subgroup. Repeatability was high across all phototypes (CV = 4.5%; provisional ICC (2, 1) = 0.975, 95% CI 0.970 - 0.979). Conclusions: CTK Derma AI provides robust, objective, and repeatable quantification of key facial aging features across diverse skin phototypes and outperforms a conventional non-calibrated baseline workflow.