A Self-guided Multimodal Approach to Enhancing Graph Representation Learning for Alzheimer's Diseases

TL;DR

The paper tackles the challenge of enhancing graph-based representations for Alzheimer's disease by leveraging uncurated, natural-language domain knowledge. It introduces a self-guided multimodal GNN (MM-GNN) that fuses brain connectome graphs with external knowledge through a backbone $f_B$ and a fusion $f_F$, and learns real-valued masks $M_d$ and $M_k$ for differentiable, soft retrieval guided by a loss $\mathcal{L}_{exp}$. External knowledge is encoded from a PubMed-derived corpus using a language model $h$ (e.g., $E_{\inom{K}$^i} = \mathrm{MLP}(h(k_i))$), forming a fusion graph $g_f$ that enables end-to-end training and graph augmentation. Experiments on OASIS and ADNI-D across DTI and fMRI modalities show improved accuracy, AUC, and F1, along with interpretable graph saliency and knowledge-wise explanations, demonstrating scalability and reduced need for expert-driven design.

Abstract

Graph neural networks (GNNs) are powerful machine learning models designed to handle irregularly structured data. However, their generic design often proves inadequate for analyzing brain connectomes in Alzheimer's Disease (AD), highlighting the need to incorporate domain knowledge for optimal performance. Infusing AD-related knowledge into GNNs is a complicated task. Existing methods typically rely on collaboration between computer scientists and domain experts, which can be both time-intensive and resource-demanding. To address these limitations, this paper presents a novel self-guided, knowledge-infused multimodal GNN that autonomously incorporates domain knowledge into the model development process. Our approach conceptualizes domain knowledge as natural language and introduces a specialized multimodal GNN capable of leveraging this uncurated knowledge to guide the learning process of the GNN, such that it can improve the model performance and strengthen the interpretability of the predictions. To evaluate our framework, we curated a comprehensive dataset of recent peer-reviewed papers on AD and integrated it with multiple real-world AD datasets. Experimental results demonstrate the ability of our method to extract relevant domain knowledge, provide graph-based explanations for AD diagnosis, and improve the overall performance of the GNN. This approach provides a more scalable and efficient alternative to inject domain knowledge for AD compared with the manual design from the domain expert, advancing both prediction accuracy and interpretability in AD diagnosis.

Figures (4)

• Figure 1: Illustration of designing AD-specific GNNs.
• Figure 2: Design overview of our approach.
• Figure 3: Brain saliency maps identified by our approach. Top 10 salient ROIs are highlighted.
• Figure 4: Distribution of importance scores of domain knowledge from our approach.