Privacy-preserving harmonization via distributed ComBat.

Area of Science:

• Neuroimaging analysis
• Medical data privacy
• Biostatistics

Background:

• Clinical data sharing is limited by privacy concerns and institutional data silos.
• Neuroimaging data integration across sites is hampered by scanner-induced biases.
• Scanner effects significantly degrade the reliability of downstream neuroimaging analyses.

Purpose of the Study:

• To develop a privacy-preserving method for harmonizing neuroimaging data across multiple institutions.
• To address the challenge of removing scanner effects in a distributed data setting.
• To enable robust downstream analyses on harmonized, multi-institutional neuroimaging datasets.

Main Methods:

• Introduced distributed ComBat, an adaptation of the ComBat harmonization technique for distributed data.
• Developed a fast and simple distributed algorithm for multivariate data harmonization.
• Validated the method using data from the Alzheimer's Disease Neuroimaging Initiative (ADNI).

Main Results:

• Distributed ComBat effectively removes scanner effects in a distributed setting.
• The algorithm yields results equivalent to traditional ComBat when applied to harmonized data.
• Maximal privacy protection is maintained throughout the harmonization process.

Conclusions:

• Distributed ComBat offers a viable solution for harmonizing neuroimaging data while preserving patient privacy.
• The method facilitates multi-institutional data integration and enhances the reliability of neuroimaging studies.
• Enables broader application of advanced analytical techniques in functional and structural neuroimaging.