TRAM-AI Checklist for Verifying Usable Medical Imaging Datasets in AI Development

A 14-item TRAM-AI checklist for verifying medical imaging dataset accessibility for AI was developed via Delphi consensus. In pilot testing of 10 datasets, compliance was 10% for archive validation and 30% for partition keys, highlighting gaps in functional usability.

• The TRAM-AI checklist contains 14 items across four domains: Dataset Access and Persistence, Technical Interoperability, Metadata and Reproducibility, and Ethical, Legal, and Maintenance considerations.
• Pilot validation in 10 publicly available datasets showed 90% compliance for trusted repository, de-identification, and subject-level mapping, but only 10% for archive validation and 30% for partition keys.
• All four advisory group members from Latin America and sub-Saharan Africa endorsed the checklist, encouraging cross-regional applicability.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.