A New Approach Based on Genetic Algorithms for Explaining Decisions in the Classification of Pathological Lesions in Medical Images

• Grant number: 25/13249-4
• Support Opportunities: Scholarships in Brazil - Master
• Start date: November 01, 2025
• End date: August 31, 2027
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Renato Tinós
• Grantee: Gabriel Branco Vitorino
• Host Institution: Faculdade de Filosofia, Ciências e Letras de Ribeirão Preto (FFCLRP). Universidade de São Paulo (USP). Ribeirão Preto , SP, Brazil
• Associated research grant: 24/08485-8 - Center for artificial intelligence in health management, AP.PCPE

Abstract

This project proposes the development of a new explainable artificial intelligence (AI) algorithm based on genetic algorithms, aimed at interpreting the decisions of convolutional neural networks applied to the classification of pathological lesions in medical images. Starting from the limitations of current explainability methods-particularly the LOcal Rule-based Explanations method, the main genetic algorithm-based approach-the project seeks to improve the quality and interpretability of explanations generated for the decisions made by black-box machine learning models in image analysis tasks. To achieve this, the project will investigate the use of segmentations based on high-level visual concepts and the generation of realistic synthetic neighborhoods using autoencoders and diffusion models. The approach will initially be applied to public datasets containing dermatoscopic images and clinical metadata associated with different types of skin lesions. Other datasets involving images of pathological lesions may also be considered. The methodology includes training predictive models with convolutional architectures, applying classical explainability methods for comparison, and developing a hybrid strategy that combines the proposed innovations in semantic segmentation, latent encoding, and genetic optimization.By making AI model decisions more transparent and interpretable, the project aims to promote their clinical acceptance and contribute to the safer and more effective use of AI in medical practice.