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

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• AI enhances clinical oncology imaging and radiotherapy.
• Growing concerns exist regarding sex-related bias in AI algorithms.
• Female-specific data is often omitted in multi-organ segmentation development.

Purpose of the Study:

• Review the importance of biological sex in AI-based multi-organ segmentation for oncology.
• Identify sources of sex-based bias in AI development and implementation.
• Provide recommendations for ensuring AI equity in this field.

Main Methods:

• Literature review focusing on AI, medical imaging, and oncology.
• Analysis of data generation, model building, and implementation processes.
• Discussion of sex as a variable in AI algorithm development.

Main Results:

• Biological sex is a critical factor influencing AI segmentation accuracy.
• Sex-based bias can arise from data collection, algorithm design, and deployment.
• Current AI models may not generalize equitably across sexes.

Conclusions:

• Addressing sex-specific data is vital for unbiased AI in oncology.
• Improved sex inclusion is necessary for fair and generalizable AI technologies.
• Recommendations are proposed to promote AI equity in imaging-guided radiotherapy.