A Multi-Dimensional Framework for Data Quality Assurance in Cancer Imaging Repositories.

Background/Objectives: Cancer remains a leading global cause of death, with breast, lung, colorectal, and prostate cancers being among the most prevalent. The integration of Artificial Intelligence (AI) into cancer imaging research offers opportunities for earlier diagnosis and personalized treatment. However, the effectiveness of AI models depends critically on the quality, standardization, and fairness of the input data. The EU-funded INCISIVE project aimed to create a federated, pan-European repository of imaging and clinical data for cancer cases, with a key objective to develop a robust framework for pre-validating data prior to its use in AI development. Methods: We propose a data validation framework to assess clinical (meta)data and imaging data across five dimensions: completeness, validity, consistency, integrity, and fairness. The framework includes procedures for deduplication, annotation verification, DICOM metadata analysis, and anonymization compliance. Results: The pre-validation process identified key data quality issues, such as missing clinical information, inconsistent formatting, and subgroup imbalances, while also demonstrating the added value of structured data entry and standardized protocols. Conclusions: This structured framework addresses common challenges in curating large-scale, multimodal medical data. By applying this approach, the INCISIVE project ensures data quality, interoperability, and equity, providing a transferable model for future health data repositories supporting AI research in oncology.