Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multi-modal Imputation for Alzheimer's Disease Classification

Abhijith Shaji, Tamoghna Chattopadhyay, Sophia I. Thomopoulos, Greg Ver Steeg, Paul M. Thompson, Jose-Luis Ambite

TL;DR

The paper tackles the challenge of incomplete multimodal MRI data for Alzheimer's disease classification by learning a conditional diffusion model to impute missing diffusion-weighted images from T1 scans. A 3D DDPM conditioned on T1 is trained to produce anatomically consistent synthetic DWI volumes, enabling larger multimodal datasets and broader evaluation of uni- and bimodal classifiers. Across extensive experiments, imputation yields improvements in several metrics, particularly for minority classes, but gains are configuration-dependent and not universally observed; in some bimodal setups, gains are largely driven by increased T1 data rather than synthetic DWI. Overall, the work demonstrates the potential and limitations of diffusion-based cross-modality synthesis for neuroimaging classification and outlines concrete directions for expanding paired datasets and exploring newer diffusion bridging methods.

Abstract

Deep learning has been successful in predicting neurodegenerative disorders, such as Alzheimer's disease, from magnetic resonance imaging (MRI). Combining multiple imaging modalities, such as T1-weighted (T1) and diffusion-weighted imaging (DWI) scans, can increase diagnostic performance. However, complete multimodal datasets are not always available. We use a conditional denoising diffusion probabilistic model to impute missing DWI scans from T1 scans. We perform extensive experiments to evaluate whether such imputation improves the accuracy of uni-modal and bi-modal deep learning models for 3-way Alzheimer's disease classification-cognitively normal, mild cognitive impairment, and Alzheimer's disease. We observe improvements in several metrics, particularly those sensitive to minority classes, for several imputation configurations.

Multi-modal Imputation for Alzheimer's Disease Classification

TL;DR

The paper tackles the challenge of incomplete multimodal MRI data for Alzheimer's disease classification by learning a conditional diffusion model to impute missing diffusion-weighted images from T1 scans. A 3D DDPM conditioned on T1 is trained to produce anatomically consistent synthetic DWI volumes, enabling larger multimodal datasets and broader evaluation of uni- and bimodal classifiers. Across extensive experiments, imputation yields improvements in several metrics, particularly for minority classes, but gains are configuration-dependent and not universally observed; in some bimodal setups, gains are largely driven by increased T1 data rather than synthetic DWI. Overall, the work demonstrates the potential and limitations of diffusion-based cross-modality synthesis for neuroimaging classification and outlines concrete directions for expanding paired datasets and exploring newer diffusion bridging methods.

Abstract

Deep learning has been successful in predicting neurodegenerative disorders, such as Alzheimer's disease, from magnetic resonance imaging (MRI). Combining multiple imaging modalities, such as T1-weighted (T1) and diffusion-weighted imaging (DWI) scans, can increase diagnostic performance. However, complete multimodal datasets are not always available. We use a conditional denoising diffusion probabilistic model to impute missing DWI scans from T1 scans. We perform extensive experiments to evaluate whether such imputation improves the accuracy of uni-modal and bi-modal deep learning models for 3-way Alzheimer's disease classification-cognitively normal, mild cognitive impairment, and Alzheimer's disease. We observe improvements in several metrics, particularly those sensitive to minority classes, for several imputation configurations.
Paper Structure (16 sections, 1 equation, 4 figures, 1 table)

This paper contains 16 sections, 1 equation, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Imputation using Image Translation
  3. Data and Preprocessing
  4. Deep Learning Architectures
  5. 3D DDPM Architecture
  6. Uni-Modal 3D CNN Architecture
  7. Bi-Modal 3D CNN Architecture
  8. Training the models
  9. 3D CNNs
  10. Conditional DDPM
  11. Results
  12. DWI-Only Model
  13. T1-Only Model
  14. Bimodal (T1 + DWI) Model
  15. Discussion
  16. ...and 1 more sections

Figures (4)

  • Figure 1: 3D UNet architecture used to predict the noise output from a noisy DWI FA MRI conditioned on T1 MRI.
  • Figure 2: T1-weighted MRI, synthetic DWI FA MRI, and real DWI FA MRI across axial, coronal, and sagittal views for a single healthy subject.
  • Figure 3: Unimodal 3D CNN Architecture.
  • Figure 4: Bimodal 3D CNN Architecture.