Framework for fairness mitigation in medical images

• Grant number: 23/12468-9
• Support Opportunities: Scholarships in Brazil - Post-Doctoral
• Start date: November 01, 2023
• End date: April 30, 2025
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computer Systems
• Agreement: Swiss National Science Foundation (SNSF)
• Principal Investigator: Lilian Berton
• Grantee: Maira Blumer Fatoretto
• Host Institution: Instituto de Ciência e Tecnologia (ICT). Universidade Federal de São Paulo (UNIFESP). Campus São José dos Campos. São José dos Campos , SP, Brazil
• Associated research grant: 21/14725-3 - FairMI: machine learning fairness with application to medical images, AP.R

Abstract

The use of Artificial Intelligence (AI) through Machine Learning (ML) in everyday life is a reality in several applications, from recommendation systems, financial systems and even in medicine. Randomized trials estimate the mean treatment effects for a population, but participants in clinical trials are often not representative of the patient population receiving treatment in terms of race and gender. As a result, medications and interventions are not tailored to historically vulnerable groups, for example, women, minority groups and obese patients tend to have generally poorer treatment options. As the example data that is used in training an ML model is produced through design and human action, it is not free from possible bias. Considering the ethical implications arising from the lack of representation in the data or even the introduction of prejudices in them, reflected in ML models, the objective of this project is to develop a framework for fairness mitigation in medical images. It will be implemented fairness measurements, baselines approaches to mitigate fairness and generate fair classification models. The developed framework will be open source available at Github. (AU)