Research Goal-Driven Data Model and Harmonization for De-Identifying Patient Data in Radiomics

Area of Science:

• Oncology
• Medical Informatics
• Data Science

Background:

• Existing de-identification systems for radiation oncology data lack flexibility in data modeling and attribute normalization.
• Defining de-identification requirements is crucial and context-dependent on specific research goals.

Purpose of the Study:

• To describe a novel de-identification process for radiation and clinical oncology data.
• To present a flexible system guided by a data model and schema for dynamic domain ontology capture and terminology normalization.

Main Methods:

• De-identification of radiological images (DICOM format) and clinical data (CSV format).
• Utilized a data model and schema for information organization, dynamic ontology capture, and terminology normalization.
• Ensured preservation of longitudinal date changes, incremental de-identification, and referential data integrity.

Main Results:

• Developed a generic model for organizing information and de-identifying clinical data.
• Successfully harmonized de-identified data across image and clinical datasets.
• Presented four case studies (glioblastoma multiforme, head-neck, breast, lung cancer) with experimental validation.

Conclusions:

• The proposed de-identification model offers a flexible and harmonized approach for radiation oncology data.
• This method supports research goals by enabling effective data sharing and analysis.
• The system ensures data integrity and facilitates the use of diverse oncology datasets in medical research.