V. Sounderajah, Ahmad Guni, Xiaoxuan Liu, G. Collins, A. Karthikesalingam, S. Markar, Robert M. Golub, A. Denniston et al.
A systematic guideline to improve the reporting quality of AI-based diagnostic accuracy studies.
AI-based diagnostic studies face difficulties in result interpretation and reproducibility due to methodological diversity and incomplete reporting. Existing STARD guidelines do not reflect AI-specific characteristics, making it difficult to ensure research transparency and reliability.
STARD-AI derived core reporting items for AI diagnostic studies through the Delphi method and expert consensus. A checklist was developed including AI-specific elements such as dataset characteristics, model development process, validation methods, and performance evaluation metrics.
Provides reporting standards for AI diagnostic studies, enhancing transparency, reproducibility, and comparability. Offers practical guidance for researchers, reviewers, and editors, and is expected to contribute to improving reliability in the field of AI medical diagnostics.