ARMARecon: An ARMA Convolutional Filter based Graph Neural Network for Neurodegenerative Dementias Classification

TL;DR

This work tackles early detection of neurodegenerative dementias from diffusion MRI by introducing ARMARecon, a graph neural network that fuses ARMA graph filtering with a reconstruction-driven regularizer. It builds a subject-level graph from histogram-based FA features across nine white-matter ROIs and employs an ARMA filter, expressed as $X = \left( I + \sum_{k=1}^{K} q_k L^k \right)^{-1} \left( \sum_{k=0}^{K-1} p_k L^k \right) H$, with $\tilde{A} = D^{-{1/2}}\hat{A}D^{-{1/2}}$, $\hat{A}=A+I_n$, to capture both local and global graph dynamics while mitigating oversmoothing. A reconstruction objective, implemented via an MLP-based decoder, enforces feature fidelity and sharpening to improve robustness under label scarcity. Comprehensive experiments on multi-site ADNI and NIFD diffusion MRI data show ARMARecon achieving superior accuracy, precision, recall, F1, and AUC compared to traditional ML models and prevailing GNNs, demonstrating strong generalization across cohorts. The approach offers scalable, interpretable, and privacy-preserving dementia classification capabilities across institutional datasets.

Abstract

Early detection of neurodegenerative diseases such as Alzheimer's Disease (AD) and Frontotemporal Dementia (FTD) is essential for reducing the risk of progression to severe disease stages. As AD and FTD propagate along white-matter regions in a global, graph-dependent manner, graph-based neural networks are well suited to capture these patterns. Hence, we introduce ARMARecon, a unified graph learning framework that integrates Autoregressive Moving Average (ARMA) graph filtering with a reconstruction-driven objective to enhance feature representation and improve classification accuracy. ARMARecon effectively models both local and global connectivity by leveraging 20-bin Fractional Anisotropy (FA) histogram features extracted from white-matter regions, while mitigating over-smoothing. Overall, ARMARecon achieves superior performance compared to state-of-the-art methods on the multi-site dMRI datasets ADNI and NIFD.

Figures (1)

• Figure 1: Proposed ARMARecon architecture integrating ARMA graph filtering with reconstruction regularization.