Fairness of an AI System in the Case of a Biobank of Images and Imaging Biomarkers

Abstract

The chapter analyzes the legal and ethical aspects of “fairness” in relation to imaging biobanks, and the application of artificial intelligence systems in developing imaging biomarkers. To talk about the fairness of AI systems in the case of a biobank of images, the report titled “FUTURE-AI: Guiding Principles and Consensus Recommendations for Trustworthy Artificial Intelligence in Medical Imaging” is analyzed. We will focus on the idea of “fairness” as defined by FUTURE-AI “For Equitable AI in Medical Imaging” and in light of the European Commission’s High-Level Expert Group on AI’s “Ethics Guidelines for Trustworthy AI”. In the case of AI training, there is a risk that trained AI algorithms will become biased toward underrepresented groups and similarly situated individuals and hence exacerbate existing health disparities. This tells us how discrimination occurs in imaging biobanks: AI tools can generate undetected errors with harmful consequences to the patient when they are applied to imaging conditions that may differ or unexpectedly deviate, even slightly, from the conditions used for training. Also, overconfidence in the outcome of AI support may affect fairness, and specific attention should be paid to diagnostic images, especially the interpretation of mammograph images, and to the skills of experienced and less experienced radiologists.

Ähnliche Arbeiten