An Automatic AI-Based Algorithm That Grades the Scalp Surface Exfoliating Process From Video Imaging. Application to Dandruff Severity and Its Validation on Subjects of Different Ages and Ethnicities

Abstract

Objectives

To evaluate the technical assets of a new imaging device that, wifi linked to a AI based algorithm, automatically grades in vivo the exfoliating process of the skin, taking dandruff as model.

Material and Methods

The hand portable device comprises a camera that possibly uses three illuminating conditions (white LED diffused lamp, cross-polarized white light and UVA rays). The learning phase of the algorithm was built on 3600 images of the vertex area of 234 subjects of different ages and three ethnicities with and without dandruff. This learning phase allowed 15 experts and dermatologists to score regarding a 6-point atlas of dandruff severities, taken as reference. In a second validation phase, 460 images from 192 subjects of different ages and ethnic background/phototypes, were automatically analyzed by the AI based device, allowing to calculate the correlation between expert's assessments and the gradings provided by the device, and, as second indicator, to compute the Mean Average Error (MAE) between both variables.

Results

The values were found significantly correlated (r² = 0.952; p < 0.001) with an overall MAE of 0.16 grading units, although presenting some differences according to ethnic background and phototypes (0.12–0.24).

Conclusion

This new imaging device coupled with AI-based analysis allows a valid, rapid, and easy determination of the scalp exfoliating process and may represent a complementary help in the diagnosis of dermatologists in some other scalp disorders. Its versatility, easy handling, and immediate AI-based analysis suggest that it may be applied to other cosmetic areas (skincare, makeup, haircare, etc.).