Normative Modeling using Multimodal Variational Autoencoders to Identify Abnormal Brain Volume Deviations in
Sayantan Kumar1,2, Philip R O Payne2, Aristeidis Sotiras2,3
1Department of Computer Science, Washington University in St Louis, St Louis, USA.
Proceedings of Spie--The International Society for Optical Engineering
|December 22, 2023
Summary
This study introduces a multi-modal deep learning model for Alzheimer's Disease (AD) research. The framework enhances brain imaging analysis by integrating multiple MRI data types, improving detection of AD-related brain changes.
Area of Science:
- Neuroimaging and Computational Neuroscience
- Artificial Intelligence in Medicine
- Alzheimer's Disease Research
Background:
- Normative modeling quantifies individual deviations from healthy brain patterns.
- Current deep learning models for normative modeling use single MRI modalities.
- Alzheimer's Disease (AD) is multifactorial, necessitating multi-modal data integration.
Purpose of the Study:
- To develop a multi-modal variational autoencoder (mmVAE) for normative modeling in AD.
- To capture joint distributions across different MRI modalities for enhanced AD analysis.
- To identify abnormal brain volume deviations associated with AD progression.
Main Methods:
- Input: Freesurfer processed brain region volumes from T1-weighted (cortical, subcortical) and T2-weighted (hippocampal) MRI scans.
- Framework: Multi-modal variational autoencoder (mmVAE) to learn healthy brain morphology.
- Application: Quantifying brain volume deviations in AD patients compared to the normative model.
Main Results:
- The mmVAE framework demonstrated higher sensitivity to AD disease staging compared to a unimodal baseline.
- mmVAE-generated deviation maps showed better correlation with patient cognitive status.
- A greater number of brain regions exhibited statistically significant deviations with mmVAE.
Conclusions:
- Multi-modal normative modeling using mmVAE effectively identifies AD-related brain abnormalities.
- Integrating multiple MRI modalities improves the sensitivity and clinical relevance of normative models for AD.
- This approach offers a more comprehensive understanding of brain heterogeneity in AD.
