Implementation of Deep Artificial Neural Networks to Support the Diagnosis of Oral Cancerous Lesions

• Grant number: 25/20570-3
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: December 01, 2025
• End date: November 30, 2026
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: João Paulo Papa
• Grantee: Julio Cesar Silva de Sousa
• Host Institution: Faculdade de Ciências (FC). Universidade Estadual Paulista (UNESP). Campus de Bauru. Bauru , SP, Brazil

Abstract

Oral cancer represents a serious public health problem in Brazil, with a high annual incidence and a still-high mortality rate, especially due to late diagnosis. Squamous cell carcinoma (SCC) is the most prevalent type, primarily affecting men over 60 years of age and affecting areas such as the tongue, floor of the mouth, and lips. Despite the simplicity of the clinical examination, failures in early identification hinder effective treatment. In this context, artificial intelligence has emerged as a diagnostic support tool, with convolutional neural networks (CNNs) and, more recently, transformer-based architectures demonstrating high performance in medical image classification tasks. This project aims to investigate and compare deep learning architectures applied to the automated classification of clinical images of oral cancer. Models based on CNNs and transformers (Vision Transformer and Swin Transformer) will be implemented and evaluated, considering binary (cancer vs. non-cancer) and multiclass (different types of oral cancer) classification tasks. Performance will be measured using metrics such as accuracy, sensitivity, specificity, F1-score, and AUC-ROC, in addition to analyzing aspects such as inference time, scalability, and interpretability for clinical application. The research will utilize a database of clinical images provided by the Bauru School of Dentistry (USP), previously anonymized and categorized according to diagnoses confirmed by anatomopathological examination. The methodological pipeline includes image preprocessing, division into training, validation, and test sets, as well as cross-validation for statistical robustness. The expected outcome is to obtain quantitative evidence on the advantages and limitations of each architecture in the early diagnosis of oral cancer, providing support for the development of clinical support systems. The project also plans to disseminate the results in journals and conferences, as well as strengthen inter-institutional collaborations, contributing to the consolidation of artificial intelligence in the context of public health dentistry. (AU)