Automated method for color reproduction in facial prosthetics using digital images and machine learning algorithms

• Grant number: 24/09151-6
• Support Opportunities: Scholarships in Brazil - Master
• Start date: May 01, 2025
• End date: April 30, 2026
• Field of knowledge: Health Sciences - Dentistry
• Principal Investigator: Cláudia Helena Lovato da Silva
• Grantee: Helena Cristina Aguiar
• Host Institution: Faculdade de Odontologia de Ribeirão Preto (FORP). Universidade de São Paulo (USP). Ribeirão Preto , SP, Brazil

Abstract

This study will evaluate the association of images captured by a cell phone with a machine learning algorithm in predicting pigments for coloring facial prostheses. With the classification of skin types defined through the Individual Typology Angle (ITA), 225 individuals will be selected and distributed evenly into 4 groups. After collecting demographic information, participants will be subjected to standardized photographs of their skin color in the "cheekbone" region. At the same time, silicone specimens will be made by weighing and combining 5 pigments to form a color palette. Each specimen will be assigned, by a specialist, to 200 participants according to their similarity to their skin color, with the aim of creating a database for artificial intelligence training. After training, photographs of 25 individuals who are not included in the algorithm's database will be provided to the software, which will grant the pigments in the proportions indicated for each skin. Then, new specimens will be made using the information from the algorithm. The specialist will also assign specimens in a conventional manner to individuals. The performance of the machine learning algorithm will be tested by calculating the mean squared error and R2 score. The specimens created using the software will be positioned on the "cheekbones" of the 25 participants and the degree of acceptance and degree of perception will be visually evaluated by 3 experts and 3 lay individuals. Furthermore, color readings of the specimens will be carried out with the spectrophotometer and the Euclidean distance values from the CIELab system between the colors provided by the algorithm and those indicated by the specialist will be calculated (Delta E). The data will be subjected to the Normality Test (Shapiro-Wilk) and hocedasticity (Levene) to define the statistical test to be used, considering p as less than 0,05 (IBM SPSS Statistics for Windows 21.0 software; IBM Corp.).