Abstract:Pooling MRI data from multiple datasets requires harmonization to reduce undesired inter-site variabilities, while preserving effects of biological variables (or covariates). The popular harmonization approach ComBat uses a mixed effect regression framework that explicitly accounts for covariate distribution differences across datasets. There is also significant interest in developing harmonization approaches based on deep neural networks (DNNs), such as conditional variational autoencoder (cVAE). However, current DNN approaches do not explicitly account for covariate distribution differences across datasets. Here, we provide mathematical results, suggesting that not accounting for covariates can lead to suboptimal harmonization outcomes. We propose two DNN-based harmonization approaches that explicitly account for covariate distribution differences across datasets: covariate VAE (coVAE) and DeepResBat. The coVAE approach is a natural extension of cVAE by concatenating covariates and site information with site– and covariate-invariant latent representations. DeepResBat adopts a residual framework inspired by ComBat. DeepResBat first removes the effects of covariates with nonlinear regression trees, followed by eliminating site differences with cVAE. Finally, covariate effects are added back to the harmonized residuals. Using three datasets from three different continents with a total of 2787 participants and 10085 anatomical T1 scans, we find that DeepResBat and coVAE outperformed ComBat, CovBat and cVAE in terms of removing dataset differences, while enhancing biological effects of interest. However, coVAE hallucinates spurious associations between anatomical MRI and covariates even when no association exists. Therefore, future studies proposing DNN-based harmonization approaches should be aware of this false positive pitfall. Overall, our results suggest that DeepResBat is an effective deep learning alternative to ComBat.

Simultaneous Dimension Reduction And Adjustment For Confounding Variation

AC-PCA Adjusts for Confounding Variation and Recovers the Anatomical Structure of Neocortex

AC-PCoA: Adjustment for Confounding Factors Using Principal Coordinate Analysis

Joint Modeling of Cellular Heterogeneity and Condition Effects with scPCA in Single-Cell RNA-Seq

Methods for Joint Imaging and RNA-seq Data Analysis

PARE: A framework for removal of confounding effects from any distance-based dimension reduction method

Joint dimension reduction and clustering analysis of single-cell RNA-seq and spatial transcriptomics data

Spatially aware dimension reduction for spatial transcriptomics

Accuracy, robustness and scalability of dimensionality reduction methods for single-cell RNA-seq analysis

High-dimensional Genomic Data Bias Correction and Data Integration Using MANCIE

Integrative, high-resolution analysis of single cells across experimental conditions with PARAFAC2

An Efficient Sufficient Dimension Reduction Method for Identifying Genetic Variants of Clinical Significance

Multiple Comparison Procedures for Neuroimaging Genomewide Association Studies

DeepResBat: deep residual batch harmonization accounting for covariate distribution differences

Dimension Reduction using Local Principal Components for Regression-based Multi-SNP Analysis in 1000 Genomes and the Canadian Longitudinal Study on Aging (CLSA)

Column subset selection for single-cell RNA-Seq clustering

Estimating and accounting for unobserved covariates in high dimensional correlated data

Deterministic column subset selection for single-cell RNA-Seq

ACE of Space: Estimating Genetic Components of High-Dimensional Imaging Data

A Comparison for Dimensionality Reduction Methods of Single-Cell RNA-seq Data

Covariance Regression with High-Dimensional Predictors