Multimodal data curation via interoperability: use cases with the Medical Imaging and Data Resource Center

Area of Science:

• Medical informatics
• Data science
• Biomedical imaging

Background:

• Interoperability is crucial for integrating multimodal datasets from diverse sources.
• The Medical Imaging and Data Resource Center (MIDRC) prioritizes data commons interoperability.
• Existing data repositories often face challenges in data sharing and integration.

Purpose of the Study:

• To demonstrate and validate the interoperability between MIDRC, BioData Catalyst (BDC), and the National Clinical Cohort Collaborative (N3C).
• To showcase the creation of integrated clinical and imaging patient cohorts from multiple data sources.
• To provide a framework for utilizing data repository interoperability for AI/ML model development.

Main Methods:

• Leveraged existing interoperability features of MIDRC, BDC, and N3C.
• Constructed two example cohorts with matched clinical and imaging data for specific use cases.
• Assessed cohort representativeness against CDC population statistics using Jensen-Shannon distance.

Main Results:

• Successfully established interoperability pathways between MIDRC, BDC, and N3C.
• Created representative multimodal patient cohorts by integrating data from the repositories.
• Validated the feasibility of using Jensen-Shannon distance for cohort representativeness analysis.

Conclusions:

• The demonstrated interoperability facilitates the creation of comprehensive multimodal datasets.
• This approach supports the development of advanced artificial intelligence and machine learning models in healthcare.
• MIDRC, BDC, and N3C users can benefit from these methods for data integration and analysis.