Protótipo baseado no aprendizado de máquina para localização anatômica do canal mandibular

Abstract

BACKGROUND: Accurate anatomical identification of the mandibular canal is a critical step in the planning of various dental interventions. It substantially reduces the risk of IAN injuries, which can result in permanent neurosensory deficits, functional implications, and a negative impact on patients' quality of life. In this scenario, the incorporation of AI-based technologies, such as machine learning systems, can represent an effective strategy for minimizing diagnostic errors. Furthermore, this project aims to significantly contribute to the scientific and professional community in the North region of Brazil, particularly in the state of Amazonas, by promoting advances in the use of technologies applied to dentistry. GENERAL OBJECTIVE: To develop a machine learning model for the automated identification of the mandibular canal on panoramic radiographs, as an initial step in building a prototype to support dental surgical planning. METHODOLOGY: This is an innovation project based on a retrospective study database. Panoramic radiographs of patients treated at the Dental Polyclinic of the Amazonas State University (UEA) with a diagnosis of impacted or semi-impacted lower third molars during the period 2020-2022 were selected. In terms of software technology and engineering, the proposed approach is to use machine learning algorithms to automatically mark/delimit the region corresponding to the inferior alveolar nerve (IAN) in images obtained through panoramic radiographs. The algorithm's output will highlight the IAN region. This delimitation will speed up treatment and reduce patient costs. RESULTS: The dataset used in this study contains 530 annotated X-ray images, distributed among training, validation, and test sets. The images and masks were adjusted using a Python script. This code converts RGB images with yellow markings into binary masks in YOLOv11 format. This project was submitted to the PAPB-CT&I/FAPEAM funding notice under number: 87711.UNI1089.2713.27032025, with the aim of obtaining financial assistance for its respective development, project approved in the contracting phase.