Observing the statistical behavior of a deep neural network during the age and sex classification task based on dental panoramic images

• Grant number: 23/15504-6
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: February 01, 2024
• End date: January 31, 2026
• Field of knowledge: Engineering - Biomedical Engineering - Medical Engineering
• Principal Investigator: Luiz Otavio Murta Junior
• Grantee: Davi Rodrigues Pultrini
• Host Institution: Faculdade de Filosofia, Ciências e Letras de Ribeirão Preto (FFCLRP). Universidade de São Paulo (USP). Ribeirão Preto , SP, Brazil

Abstract

Forensic dentistry aims to identify humans using parts of the oral cavity in situations of death in disasters. These techniques are highly efficient when antemortem information is available that allows comparison with postmortem data. However, when such antemortem information is not available, secondary methods are considered to determine information that at least characterizes the individual (age, sex, ancestry, eye color, hair, etc.). Panoramic radiography (PAN) is an imaging technique widely used for this purpose, as it allows a detailed analysis of the dental structure and teeth, which has recently been combined with Artificial Intelligence (AI) for a faster and more accurate analysis. However, current AI applications with PAN images have focused on the segmentation, classification and numbering of teeth for age estimation, for example. However, these methods tend to exhibit limitations when approaching adult subjects for whom the dental development process has already been completed. In this project, we are proposing the development of Artificial Intelligence models for estimating Age and determining Sex from PAN Images not based on dental analysis, but on the entire PAN image. Furthermore, through the AI already built, it is possible to study the LMC statistical complexity of the model used through mathematical and physical studies, defined by the product of Shannon entropy and the statistical distance from equilibrium. With this study, we intend to understand better what happens internally in the network during DL while dealing with medical images.