Avaliação do dimorfismo sexual e idade utilizando dados morfológicos da mandíbula e aprendizado de máquina

Abstract

Panoramic radiographs provide useful information for forensic anthropology and criminal investigation. Artificial Intelligence as a tool to improve forensic dentistry has the objective of optimizing and facilitating the process of identifying people. Thus, the aim of this study is to estimate age and sexual dimorphism through morphological analysis of the mandible using panoramic radiography and machine learning. This is a quantitative, cross-sectional and retrospective study. Three hundred digital panoramic radiographs obtained from the database of the Radiology and Imaginology sector of the Dental School Clinic of the Department of Dentistry – Campus I of the State University of Paraíba were used. An observer, blinded to gender and age data, analyzed the images by categorizing the position of the mental foramen in relation to the first and second premolars. Subsequently, bilateral linear and angular measurements of the mental foramen, mandibular canal, mandibular ramus and mandibular condyle were performed using the ImageJ program. The obtained data were sent to machine learning to extract the semantic attributes. The evaluation method used was cross-validation, in which the sample was divided into subgroups, with 85% of the sample in the training/validation subgroup and 15% in the test group. The machine learning models with the best results were the AdaBoost Regressor for age and the Logistic Regression for gender. It was found for MAE age determination (Mean Absolute Error) of 4.1550 years considering the entire mandible. The data obtained from the analyzes of the left side showed MAE inferior to the right side and total mandible (4.1243). For sex determination, the data obtained from the mandible showed an accuracy of 90%, AUC (Area Under ROC Curve) of 93.57% and F1 (F1 Score) of 93.02%. The data obtained for the right side were more accurate in determining sex with an accuracy of 88.33%, AUC of 94.94% and F1 of 91.95. It is concluded that morphological data of the mandible obtained through panoramic radiography and analyzed through machine learning contribute positively to the determination of sex and age in the process of identifying people.