Multimodal MRI Accurately Identifies Amyloid Status in Unbalanced Cohorts in Alzheimer's Disease Continuum

Giorgio Dolci,Charles A. Ellis,Federica Cruciani,Lorenza Brusini,Anees Abrol,Ilaria Boscolo Galazzo,Gloria Menegaz,Vince D. Calhoun

2024-10-15

Abstract:Amyloid-$\beta$ (A$\beta$) plaques in conjunction with hyperphosphorylated tau proteins in the form of neurofibrillary tangles are the two neuropathological hallmarks of Alzheimer's disease. It is well-known that the identification of individuals with A$\beta$ positivity could enable early diagnosis. In this work, we aim at capturing the A$\beta$ positivity status in an unbalanced cohort enclosing subjects at different disease stages, exploiting the underlying structural and connectivity disease-induced modulations as revealed by structural, functional, and diffusion MRI. Of note, due to the unbalanced cohort, the outcomes may be guided by those factors rather than amyloid accumulation. The partial views provided by each modality are integrated in the model allowing to take full advantage of their complementarity in encoding the effects of the A$\beta$ accumulation, leading to an accuracy of $0.762\pm0.04$. The specificity of the information brought by each modality is assessed by \textit{post-hoc} explainability analysis (guided backpropagation), highlighting the underlying structural and functional changes. Noteworthy, well-established biomarker key regions related to A$\beta$ deposition could be identified by all modalities, including the hippocampus, thalamus, precuneus, and cingulate gyrus, witnessing in favor of the reliability of the method as well as its potential in shading light on modality-specific possibly unknown A$\beta$ deposition signatures.

Artificial Intelligence

What problem does this paper attempt to address?

### Problems the Paper Attempts to Solve

This paper aims to address the issue of identifying amyloid-beta (Aβ) positive status in individuals at different stages of the Alzheimer's disease (AD) continuum. Specifically, the authors seek to accurately identify the Aβ positive status of individuals using multimodal information provided by structural magnetic resonance imaging (sMRI), resting-state functional magnetic resonance imaging (rs-fMRI), and diffusion magnetic resonance imaging (dMRI) in an imbalanced cohort. Due to the cohort imbalance, the study results may be influenced by factors other than simple amyloid-beta accumulation. Therefore, the authors propose a multimodal and interpretable deep learning framework to fully utilize the complementarity between different modalities and to assess the specificity of the information provided by each modality through post-hoc interpretability analysis (e.g., guided backpropagation).

### Main Research Objectives

1. **Capture Aβ Positive Status**: Utilize multimodal MRI data to capture the Aβ positive status of individuals at different disease stages.
2. **Handle Imbalanced Cohort**: Perform classification in an imbalanced cohort to ensure the reliability and accuracy of the results.
3. **Integrate Multimodal Information**: Combine information provided by structural, functional, and diffusion MRI to improve classification performance.
4. **Interpretability Analysis**: Evaluate the contribution of each modality to the final classification result through post-hoc interpretability methods (e.g., guided backpropagation), revealing specific brain region changes associated with Aβ deposition.

### Research Background

The pathological features of Alzheimer's disease include amyloid-beta (Aβ) plaques and neurofibrillary tangles formed by hyperphosphorylated tau protein in the brain. Early identification of Aβ positive individuals aids in the early diagnosis of the disease. Currently, the detection of Aβ positivity mainly relies on amyloid positron emission tomography (PET) and cerebrospinal fluid (CSF) tests, but these methods have limitations such as high cost, radiation exposure, and invasiveness. Therefore, utilizing non-invasive MRI techniques, especially multimodal MRI, has become a promising alternative.

### Method Overview

1. **Dataset**: Use sMRI, rs-fMRI, and dMRI data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database, covering individuals at different disease stages.
2. **Preprocessing and Feature Extraction**: Preprocess sMRI, rs-fMRI, and dMRI data to extract structural, functional, and diffusion connectome features.
3. **Deep Learning Framework**:
- **Feature Extraction Module**: Use 3D-CNN and Graph Convolutional Network (GCN) to extract latent representations of sMRI, rs-fMRI, and dMRI, respectively.
- **Data Fusion and Classification Module**: Fuse the latent features of different modalities and use a Multilayer Perceptron (MLP) for final classification.
4. **Post-hoc Interpretability Analysis**: Generate attribution maps using the guided backpropagation (GBP) method to evaluate the contribution of each modality to the classification result.

### Experimental Results

1. **Classification Performance**: The proposed multimodal framework achieved high accuracy (0.762±0.04) in the Aβ status classification task, outperforming single-modality models.
2. **Attribution Map Analysis**: Attribution maps showed that regions such as the hippocampus, thalamus, cingulate gyrus, and precuneus play important roles in the classification of Aβ positive individuals.
3. **Statistical Analysis**: Statistical analysis of important features revealed significant differences in these regions in Aβ positive individuals.

### Conclusion

This study successfully utilized multimodal MRI data to accurately identify Aβ positive status in an imbalanced cohort and revealed specific brain region changes associated with Aβ deposition through post-hoc interpretability analysis. This provides new tools and methods for the early diagnosis of Alzheimer's disease.