ComBatLS: A location- and scale-preserving method for multi-site image harmonization.

Area of Science:

• Neuroimaging
• Biostatistics
• Computational Neuroscience

Background:

• Massive datasets and harmonization methods are used for neuroanatomical normative models.
• Current tools fail to preserve biological covariate effects (sex, age), causing errors in normative scores, especially with unequal site distributions.

Purpose of the Study:

• Introduce ComBatLS, an extension of ComBat, to preserve biological variance in neuroanatomical features.
• Evaluate ComBatLS's performance in replicating normative scores and reducing sex biases.
• Assess ComBatLS's effectiveness in harmonizing large-scale, multi-study neuroimaging data.

Main Methods:

• Developed ComBatLS, a novel extension of the ComBat harmonization algorithm.
• Utilized UK Biobank data for validation, creating sex-imbalanced synthetic sites.
• Applied ComBatLS to harmonize data from over 50 studies.

Main Results:

• ComBatLS robustly replicates normative scores better than other ComBat methods in sex-imbalanced scenarios.
• ComBatLS significantly reduces sex biases in normative scores compared to traditional methods.
• ComBatLS successfully harmonizes large-scale consortium neuroimaging data.

Conclusions:

• ComBatLS effectively preserves biological variance, improving the accuracy of neuroanatomical normative models.
• This method enhances the reliability of individual benchmarking and reduces systematic biases in neuroimaging studies.
• ComBatLS offers a robust solution for harmonizing diverse neuroimaging datasets, facilitating multi-site research.