Multi-Template Meta-Information Regularized Network for Alzheimer's Disease Diagnosis Using Structural MRI

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Structural magnetic resonance imaging (sMRI) is crucial for Alzheimer's disease (AD) diagnosis, offering detailed in vivo brain morphometry.
• Existing methods inadequately address the complex influence of metadata (age, gender, education) on sMRI-based AD diagnosis, leading to potential confounding effects and biases.

Purpose of the Study:

• To develop a novel Multi-template Meta-information Regularized Network (MMRN) for more accurate computer-aided Alzheimer's disease diagnosis.
• To effectively learn and disentangle confounding metadata from disease-related features in sMRI data.

Main Methods:

• Employed multi-template spatial transformations for data augmentation in self-supervised learning.
• Integrated weakly supervised meta-information learning and mutual information minimization to regularize diagnostic representations.
• Validated the MMRN on large-scale Alzheimer's Disease Neuroimaging Initiative (ADNI) and National Alzheimer's Coordinating Center (NACC) cohorts.

Main Results:

• The MMRN significantly outperformed state-of-the-art methods in Alzheimer's disease diagnosis.
• Achieved superior performance in predicting mild cognitive impairment (MCI) conversion.
• Demonstrated high accuracy in classifying normal control (NC), MCI, and AD stages.

Conclusions:

• The proposed MMRN effectively leverages metadata for enhanced Alzheimer's disease diagnosis and prediction.
• The method's ability to disentangle confounding metadata represents a significant advancement in neuroimaging-based disease diagnostics.