Beyond Fitzpatrick: automated artificial intelligence-based skin tone analysis in dermatological patients

TL;DR

This paper presents an AI-based method to automatically analyze skin tone, improving on the limited Fitzpatrick scale for better dermatological applications.

Contribution

The study introduces an algorithm using ITA values from CIELAB color space for automated skin tone classification.

Findings

The algorithm shows high agreement with Monk classifications but less consistency with Fitzpatrick types.

It reliably classifies skin tone to the Monk scale despite class imbalance.

The method has potential for use in teledermatology and personalized medicine.

Abstract

Human skin tone is influenced by genetic, environmental, and cultural factors and plays a key role in dermatology due to variation in disease presentation across skin tones. The widely used Fitzpatrick scale, based on UV response, classifies only a small number of skin types, limiting its ability to capture the full diversity of skin tones. This study introduces an algorithm for automated skin tone assessment by calculating the Individual Typology Angle (ITA) from CIELAB color values using DensePose and OpenFace. ITA values are mapped to both Fitzpatrick and Monk skin tone scales. Validation on 3D body scans and AI-generated images showed high agreement with Monk classifications but less consistent alignment with Fitzpatrick types. Despite class imbalance, the algorithm reliably classifies skin tone to the Monk scale and holds potential for applications in teledermatology, clinical research, and personalized medicine. Further research is warranted to externally validate our algorithm.

Figures