AID-RT: Standardising Artificial Intelligence Documentation in RadioTherapy with a domain-specific model card

Area of Science:

• Radiotherapy
• Artificial Intelligence
• Medical Imaging

Background:

• Insufficient documentation of artificial intelligence (AI) models hinders reproducibility and clinical integration.
• A standardized framework is crucial for AI in radiotherapy (RT).

Purpose of the Study:

• To develop a standardized, structured, and domain-specific reporting framework for AI models in radiotherapy (RT).
• Enhance transparency and accountability in AI applications within RT.

Main Methods:

• A working group of 16 experts from 13 institutions was formed.
• Reviewed existing AI and data reporting initiatives.
• Drafted and refined a template through five review rounds and online discussions, focusing on synthetic CT, segmentation, and dose prediction tasks.

Main Results:

• A 6-section template was developed, covering metadata, model information, technical specifications, training data, evaluation data, and ethical considerations.
• The template is available as a downloadable document and an interactive web form.
• The framework is tailored for AI models in RT, applicable in research and clinical settings.

Conclusions:

• A practical, consensus-driven template for AI in RT was proposed.
• The framework conveys essential information for informed use of AI models.
• Enhances transparency and accountability for AI in radiotherapy.