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

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Cancer is a leading cause of death globally, necessitating advancements in diagnosis and treatment.
• Artificial Intelligence (AI) integration in cancer imaging holds promise for early detection and personalized medicine.
• AI model performance relies heavily on the quality, standardization, and fairness of input data.

Purpose of the Study:

• To develop a robust framework for pre-validating imaging and clinical data for AI development.
• To create a federated, pan-European repository of cancer imaging and clinical data (INCISIVE project).
• To ensure data quality, interoperability, and equity in health data repositories.

Main Methods:

• A data validation framework assessing clinical (meta)data and imaging data across five dimensions: completeness, validity, consistency, integrity, and fairness.
• Procedures included deduplication, annotation verification, DICOM metadata analysis, and anonymization compliance.
• Framework applied to data within the INCISIVE project.

Main Results:

• Identified critical data quality issues, including missing clinical information and inconsistent formatting.
• Detected subgroup imbalances in the dataset.
• Demonstrated the benefits of structured data entry and standardized protocols for data quality.

Conclusions:

• The structured framework effectively addresses challenges in curating large-scale, multimodal medical data.
• The INCISIVE project's approach ensures data quality, interoperability, and equity.
• This framework offers a transferable model for future health data repositories supporting AI research in oncology.