MLC-GCN: Multi-Level Generated Connectome Based GCN for AD Analysis

TL;DR

The paper tackles early Alzheimer's disease detection from rsfMRI by learning multi-scale brain connectivity representations. It introduces MLC-GCN, which uses a multi-graph generator with a spatio-temporal feature extraction (STFE) module to produce level-specific connectomes, each fed into its own GCN encoder; the level-wise embeddings are concatenated for final classification. A key contribution is the intra-group loss that encourages diverse, clinically meaningful connectomes across levels, leading to improved performance on ADNI and OASIS-3 with strong explainability via identified ROIs. The method achieves state-of-the-art results for multi-class AD prediction and offers interpretable connectome patterns, illustrating potential applicability to other neurological conditions and clinical outcomes.

Abstract

Alzheimer's Disease (AD) is a currently incurable neurodegeneartive disease. Accurately detecting AD, especially in the early stage, represents a high research priority. AD is characterized by progressive cognitive impairments that are related to alterations in brain functional connectivity (FC). Based on this association, many studies have been published over the decades using FC and machine learning to differentiate AD from healthy aging. The most recent development in this detection method highlights the use of graph neural network (GNN) as the brain functionality analysis. In this paper, we proposed a stack of spatio-temporal feature extraction and graph generation based AD classification model using resting state fMRI. The proposed multi-level generated connectome (MLC) based graph convolutional network (GCN) (MLC-GCN) contains a multi-graph generation block and a GCN prediction block. The multi-graph generation block consists of a hierarchy of spatio-temporal feature extraction layers for extracting spatio-temporal rsfMRI features at different depths and building the corresponding connectomes. The GCN prediction block takes the learned multi-level connectomes to build and optimize GCNs at each level and concatenates the learned graphical features as the final predicting features for AD classification. Through independent cohort validations, MLC-GCN shows better performance for differentiating MCI, AD, and normal aging than state-of-art GCN and rsfMRI based AD classifiers. The proposed MLC-GCN also showed high explainability in terms of learning clinically reasonable connectome node and connectivity features from two independent datasets. While we only tested MLC-GCN on AD, the basic rsfMRI-based multi-level learned GCN based outcome prediction strategy is valid for other diseases or clinical outcomes.

Figures (7)

• Figure 1: The overall workflow of the MLC-GCN contains: a data preprocessing module (a), a multi-graph generator (b), and a multi-level GCN-based predictor. In module a, a brain atlas is used to extract n time series from n brain regions. In b, temporal embedding and STFE are used to extract sparse temporal features at different hierarchy and to calculate the corresponding graphs (connectomes). In c, GCNs are used to encode the generated graphs at different levels into higher level graphical data features. These features are concatenated and input to a multi-layer perceptron (MLP) to classify AD.
• Figure 2: The framework of the proposed STFE module.
• Figure 3: Ablation study of MLC-GCN with 5-fold cross validation by 5 evaluation measurements.
• Figure 4: Ablation study of MLC_GCN$_6$ of selected feature levels with 5-fold cross validation. The subscript indicates the index of the feature level with "1" the initial STFE level and "6" the deepest one.
• Figure 5: Ablation study of different numbers of embedding length $l$ with 5-fold cross validation.