A deep learning-based multisite neuroimage harmonization framework established with traveling-subject dataset

Abstract

The accumulation of multisite large-sample MRI datasets collected by large brain research projects in the last decade has provided a critical resource for understanding the neurobiological mechanisms underlying cognitive functions and brain disorders. However, the significant site effects, observed in the imaging data and their derived structural and functional features, has prevented the derivation of consistent findings across different studies. The development of harmonization methods that can effectively eliminate complex site effects while maintaining biological characteristics in neuroimaging data has become a vital and urgent requirement for multisite imaging studies. Here, we proposed a deep learning-based framework to harmonize imaging data from pairs of sites, in which site factors and brain features can be disentangled and encoded. We trained the proposed framework with a publicly available traveling-subject dataset from SRPBS and harmonized the gray matter volume maps from eight source sites to a target site. The proposed framework significantly eliminated inter-site differences in gray matter volume. The embedded encoders successfully captured both the abstract texture of site factors and the concrete brain features. Moreover, the proposed framework exhibited outstanding performance relative to conventional statistical harmonization methods in site effect removal, data distribution homogenization, and intra-subject similarity improvement. Together, the proposed method offers a powerful and interpretable deep learning-based harmonization framework for multisite neuroimaging data that could enhance reliability and reproducibility in multisite studies for brain development and brain disorders.