A New Multiple Imputation Method for High-Dimensional Neuroimaging Data

Tong Lu ¹, Peter Kochunov ², Chixiang Chen ^3,4

⁰Department of Mathematics, University of Maryland, College Park, Maryland, USA.

Human Brain Mapping +

|

March 21, 2025

View abstract on PubMed

Summary

High dimensional Multiple Imputation (HIMA) offers a computationally efficient Bayesian approach to handle missing neuroimaging data. This novel method significantly reduces imputation time and improves data precision for complex brain imaging analyses.

Area Of Science

Neuroimaging
Statistical analysis
Computational statistics

Background

Missing data are a significant challenge in neuroimaging, potentially introducing bias and affecting statistical analysis.
Traditional multiple imputation methods are computationally intensive and impractical for high-dimensional neuroimaging datasets.

Purpose Of The Study

To introduce High dimensional Multiple Imputation (HIMA), a novel Bayesian approach for handling missing data in large-scale neuroimaging.
To address the computational challenges associated with multiple imputation in high-dimensional neuroimaging data.

Main Methods

HIMA utilizes Bayesian models tailored for large-scale neuroimaging datasets.
A new computational strategy samples large covariance matrices using a robustly estimated posterior mode.
The approach was validated through extensive simulation studies and real-data analysis on a schizophrenia brain imaging dataset.

Main Results

HIMA demonstrates a substantial reduction in computational burden, decreasing processing time from 800 hours (classic methods) to 1 hour.
The method significantly improves computational efficiency and numerical stability for neuroimaging data.
HIMA enhances the precision and stability of imputed data compared to traditional techniques.

Conclusions

HIMA provides an effective and computationally efficient solution for addressing missing data in high-dimensional neuroimaging.
The proposed method overcomes the limitations of existing multiple imputation techniques for neuroimaging applications.
HIMA facilitates more reliable and stable statistical inferences from neuroimaging studies with missing data.

Keywords:

Bayesian large covariance matrix multiple imputation multivariate missing data posterior mode