Using self-supervised learning to predict recurrence and malignant transformation of oral leukoplakia stratified according to type of treatment

• Grant number: 24/18766-4
• Support Opportunities: Scholarships abroad - Research Internship - Doctorate
• Start date: June 09, 2025
• End date: June 08, 2026
• Field of knowledge: Health Sciences - Dentistry
• Principal Investigator: Marcio Ajudarte Lopes
• Grantee: Caique Mariano Pedroso
• Supervisor: Alexander Thomas Pearson
• Host Institution: Faculdade de Odontologia de Piracicaba (FOP). Universidade Estadual de Campinas (UNICAMP). Piracicaba , SP, Brazil
• Institution abroad: University of Chicago, United States
• Associated to the scholarship: 22/07276-0 - Clinicopathological and digital predictors of recurrence and malignancy of oral leukoplakia and proliferative verrucous leukoplakia: a clinical trial associated with the use of artificial intelligence, BP.DR

Abstract

This project aims to evaluate a deep learning (DL) model based on self-supervised learning to predict the recurrence and malignant transformation of oral leukoplakia, stratified by surgical treatment. The study will utilize histopathological images and treatment data from patients with oral leukoplakia. Whole-slide images (WSIs) will be processed using the Slideflow tool to annotate Regions of Interest (ROI) and extract tiles for analysis. Initially, a self-supervised learning model will be pre-trained on large amounts of unlabeled data to learn robust representations of histopathological images. Subsequently, the model will be refined using a supervised approach, labeling each tile based on the diagnosis from the original slide. During training, data augmentation and stain normalization techniques will be employed to enhance model robustness. Final predictions will be based on the average of all tile predictions from each slide. The model's performance will be evaluated using metrics such as accuracy, sensitivity, specificity, F1-score, and area under curve (AUC-ROC). Statistical analyses will compare predictive results between patients treated with a scalpel versus those treated with a laser. The time to recurrence and malignant transformation will be compared between the treatment groups using Kaplan-Meier curves, with statistical significance tested using the log-rank test. This model has the potential to contribute to personalized treatment approaches for leukoplakia, assisting in the selection of the most appropriate treatment method (scalpel or laser) for each patient.