Sex-Based Bias in Artificial Intelligence-Based Segmentation Models in Clinical Oncology

Abstract

Artificial intelligence (AI) advancements have accelerated applications of imaging in clinical oncology, especially in revolutionizing the safe and accurate delivery of state-of-the-art imaging-guided radiotherapy techniques. However, concerns are growing over the potential for sex-related bias and the omission of female-specific data in multi-organ segmentation algorithm development pipelines. Opportunities exist for addressing sex-specific data as a source of bias, and improving sex inclusion to adequately inform the development of AI-based technologies to ensure their fairness, generalizability and equitable distribution. The goal of this review is to discuss the importance of biological sex for AI-based multi-organ image segmentation in routine clinical and radiation oncology; sources of sex-based bias in data generation, model building and implementation and recommendations to ensure AI equity in this rapidly evolving domain.