Skin Phototype Classification with Machine Learning Based on Broadband Optical Measurements

Area of Science:

• Biomedical Optics
• Dermatology
• Machine Learning in Healthcare

Background:

• The Fitzpatrick Skin Phototype Classification (FSPC) scale is a standard but has limitations including underrepresentation of darker skin tones and subjectivity.
• These limitations can lead to disparities in dermatological care, misdiagnosis of wound healing, and underestimation of disease severity for individuals with darker skin.
• Objective and high-resolution skin typing methods are needed to address these disparities.

Purpose of the Study:

• To develop and validate an objective method for skin phototype classification using optical sensing and machine learning.
• To compare the performance of the developed algorithm against the traditional FSPC scale.
• To explore optimization of the method across different spectral bands for clinical applications.

Main Methods:

• Development of an optical sensor measuring light reflectance from 410-940 nm.
• Application of an unsupervised K-means clustering algorithm for skin phototype classification using broadband optical data.
• Comparison of algorithm-based classification with human FSPC assessment in a diverse cohort (n=30).

Main Results:

• The FSPC scale showed limited differentiation between closely related skin phototypes (e.g., I vs. II) but could distinguish broad ranges (e.g., I vs. VI).
• The K-means algorithm demonstrated superior differentiation across a wider range of skin phototypes and wavelengths.
• The optical sensor and algorithm approach provided better classification resolution, proving quantifiable and reproducible.

Conclusions:

• An optical sensor combined with a K-means algorithm offers a more objective, reproducible, and higher-resolution method for skin phototype classification than the FSPC scale.
• This technology has the potential to mitigate dermatological care disparities by providing accurate skin typing across the full spectrum of skin tones.
• Further optimization for specific spectral bandwidths can enhance clinical utility and diagnostic accuracy in dermatology.