Imputation-Based Harmonization Mitigates Site Effects Without Data Leakage in Machine Learning Studies
Neuroimaging studies that pool data across clinical sites often suffer from site effects --- variability in imaging measures that arises from technical heterogeneity across sites as opposed to true biological signal. While numerous harmonization methods have been proposed to remove site effects from imaging features, less attention has been placed on how to incorporate harmonization models into…
Neuroimaging studies that combine data from multiple clinical sites often encounter site effects - variations in imaging measures that stem from technical differences between sites, as opposed to genuine biological signals. Although various harmonization techniques have been developed to eliminate site effects from imaging features, there has been relatively little focus on integrating these harmonization models into typical machine-learning workflows.
In their investigation, the researchers have found that existing methods for combining data either lead to data leakage, fail to completely remove site effects from the target data, or reduce the accuracy of true biological associations during the harmonization procedure.
To tackle these challenges, the researchers have introduced a novel method called Multiple Imputation for Removing Technical Heterogeneity (MIRTH). This method employs imputed outcomes to harmonize the test data, thereby addressing the issues that plague existing integration approaches. The performance of MIRTH was evaluated using both simulated and real-world volumetric data obtained from the Alzheimer's Disease Neuroimaging Initiative and the Baltimore Longitudinal Study of Aging.
The findings indicate that MIRTH can achieve high predictive accuracy while preventing inflated performance when the outcome is imbalanced across clinical sites.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
