Batch Effect Correction for Neuroimaging Data with Heterogeneous Spatial Correlations.

Area of Science:

• Neuroimaging
• Brain function analysis
• Data science

Background:

• Magnetic resonance imaging (MRI) is crucial for studying brain structure and function.
• Multi-site neuroimaging studies offer enhanced sample diversity and statistical power.
• Batch effects from varying imaging protocols introduce non-biological variability.

Purpose of the Study:

• To develop novel methods for correcting batch effects in neuroimaging data.
• To account for spatial correlations within brain images affected by multi-site data acquisition.
• To improve the reliability and accuracy of neuroimaging analyses.

Main Methods:

• Development of Covariance-Aware Multivariate (CAM) ComBat to handle spatial correlations and heterogeneous features across batches.
• Introduction of Spatially-Informed Iterative Block (SIB) ComBat as a computationally efficient alternative for high-dimensional data.
• Validation through simulation studies and application to real neuroimaging datasets.

Main Results:

• CAM-ComBat effectively accounts for spatial correlations in high-dimensional neuroimaging data.
• SIB-ComBat provides a scalable and efficient solution for large-scale neuroimaging datasets.
• Both methods demonstrate superior performance compared to existing batch effect correction techniques.

Conclusions:

• The developed CAM-ComBat and SIB-ComBat methods offer significant improvements in neuroimaging data analysis.
• These methods enhance the ability to study brain mechanisms and cognitive associations by mitigating batch effects.
• The findings support the use of these advanced techniques in large-scale, multi-site neuroimaging research.