Agnostic Computer Vision-Based explainability applied to medical images.

• Grant number: 25/06148-7
• Support Opportunities: Scholarships abroad - Research Internship - Scientific Initiation
• Start date: July 15, 2025
• End date: November 14, 2025
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: André Carlos Ponce de Leon Ferreira de Carvalho
• Grantee: Ayrton da Costa Ganem Filho
• Supervisor: Thierry Urruty
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Institution abroad: Université de Poitiers, France
• Associated to the scholarship: 24/13315-4 - Computer vision in the analysis of lip cancer: Increasing diagnostic reliability with the use of Explainable Artificial Intelligence., BP.IC

Abstract

The advancement of Computer Vision techniques, especially with Convolutional Neural Networks (CNNs), has enabled faster and more accurate medical diagnoses. However, the lack of interpretability of these models limits their clinical adoption due to a lack of transparency - particularly critical in the diagnosis of head and neck cancer. This project proposes the development and application of agnostic Explainable Artificial Intelligence (XAI) models for medical imaging, aiming to compare different techniques and various Computer Vision architectures to enhance the transparency and trust in clinical decision-making. Techniques such as Grad-CAM, LIME, SHAP, and LRP will be applied and compared both quantitatively and qualitatively using clinical photograph datasets from Brazil and the Université de Poitiers. The project aims to investigate the generalizability of the proposed techniques across different cancer types and image formats, strengthening their versatility, scalability, and, consequently, their clinical applicability in Brazil. The international collaboration will provide access to advanced infrastructure - including both datasets and analytical equipment - with considerable impact on the ongoing project and real-world applicability, making foreign group's input crucial. (AU)