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

TL;DR

This study tackles noninvasive identification of amyloid-β status across the Alzheimer's disease continuum using a multimodal MRI framework that combines structural (sMRIGM), functional (rs-fMRI FNC), and diffusion (dMRI SC) information. A two-module deep learning architecture—with modality-specific feature reduction (3D-CNN for sMRI; GCNs for rs-fMRI and dMRI) and a fusion-based MLP classifier—achieves a mean accuracy of $0.762 \\pm 0.04$, outperforming unimodal baselines. Post-hoc guided backpropagation reveals interpretable brain regions driving the predictions, including the precuneus, hippocampus, thalamus, and cingulate gyrus, highlighting cross-modality signatures and modality-specific contributions. The work demonstrates the potential of noninvasive MRI to identify amyloid status and provides a framework for explainable multimodal neuroimaging analyses, though it notes limitations related to small, unbalanced samples and calls for larger balanced cohorts to disentangle amyloid effects from disease progression.

Abstract

Amyloid-$β$ (A$β$) 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$β$ positivity could enable early diagnosis. In this work, we aim at capturing the A$β$ 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$β$ 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$β$ 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$β$ deposition signatures.

Figures (3)

• Figure 1: Schematic representation of the proposed framework. The model takes as input three MRI neuroimaging modalities: sMRI 3D volumes, rs-fMRI functional graph, and dMRI structural graph. The DL architecture is composed of two modules: i) a feature reduction module, where the input data are transformed in their latent representations; and ii) a data fusion & Classification module, where the latent feature of each modality are concatenated together and, finally, they are classified using a MLP.
• Figure 2: GBP-based attributions for the A$\beta$ positive mean subject derived from the correctly classified individuals overlaid to the MNI152 template, where: A. Saggital, coronal, and axial views for the average sMRI GBP map where only the attributions exceeding the $96^{th}$ percentile are shown, highlighting both cortical and subcortical regions; B. The $10$ most important nodes (ICs) from the rs-fMRI data, representing mainly the DM and CC brain networks; C. The $10$ most important nodes (ROIs) from the dMRI data, involving both cortical and subcortical regions in both hemispheres, also including the cerebellum.
• Figure 3: Distributions of the input data for the 10 most important brain regions/nodes considered in the statistical analysis for each modality. A. Between-subject distribution of the regional (mean) GM volumes for sMRI; B. Between-subjects distribution of the node strength values for rs-fMRI; and C. Between-subject distribution of the betweenness centrality values for dMRI.