Variational Bayesian partially linear mean shift models for high-dimensional Alzheimer's disease neuroimaging data

Area of Science:

• Neuroimaging
• Biostatistics
• Machine Learning

Background:

• Alzheimer's disease diagnosis relies on brain imaging (MRI) and cognitive tests (MMSE).
• Analyzing high-dimensional MRI data with traditional Bayesian methods for partially linear mean shift models (PLMSM) is computationally intensive and slow.
• Existing methods struggle with the computational demands and memory requirements of high-dimensional data in PLMSM.

Purpose of the Study:

• To develop an efficient computational method for analyzing the relationship between MRI data and cognitive scores in Alzheimer's disease.
• To address the limitations of existing Bayesian approaches for high-dimensional data in PLMSM.
• To simultaneously estimate parameters, nonparametric functions, and identify outliers in PLMSM for Alzheimer's research.

Main Methods:

• A variational Bayesian inference framework is proposed for PLMSM.
• Bayesian P-splines are used to approximate nonparametric functions.
• A Bayesian adaptive Lasso method is employed for predictor selection, and outliers are identified using a classification variable.

Main Results:

• The proposed variational Bayesian method offers improved computational efficiency and memory usage compared to traditional methods.
• The method effectively estimates parameters and nonparametric functions in high-dimensional settings.
• Simulation studies demonstrate the finite sample performance of the developed approach.

Conclusions:

• The novel variational Bayesian inference method provides an efficient and effective tool for analyzing high-dimensional MRI data in the context of Alzheimer's disease.
• This approach facilitates a more robust investigation of the relationship between brain imaging features and cognitive function.
• The method's ability to identify outliers aids in understanding disease heterogeneity and improving diagnostic accuracy.