Disease Prediction using Graph Convolutional Networks: Application to Autism Spectrum Disorder and Alzheimer's Disease

Abstract: Graphs are widely used as a natural framework that captures interactions between individual elements represented as nodes in a graph. In medical applications, specifically, nodes can represent individuals within a potentially large population (patients or healthy controls) accompanied by a set of features, while the graph edges incorporate associations between subjects in an intuitive manner. This representation allows to incorporate the wealth of imaging and non-imaging information as well as individual subject features simultaneously in disease classification tasks. Previous graph-based approaches for supervised or unsupervised learning in the context of disease prediction solely focus on pairwise similarities between subjects, disregarding individual characteristics and features, or rather rely on subject-specific imaging feature vectors and fail to model interactions between them. In this paper, we present a thorough evaluation of a generic framework that leverages both imaging and non-imaging information and can be used for brain analysis in large populations. This framework exploits Graph Convolutional Networks (GCNs) and involves representing populations as a sparse graph, where its nodes are associated with imaging-based feature vectors, while phenotypic information is integrated as edge weights. The extensive evaluation explores the effect of each individual component of this framework on disease prediction performance and further compares it to different baselines. The framework performance is tested on two large datasets with diverse underlying data, ABIDE and ADNI, for the prediction of Autism Spectrum Disorder and conversion to Alzheimer's disease, respectively. Our analysis shows that our novel framework can improve over state-of-the-art results on both databases, with 70.4% classification accuracy for ABIDE and 80.0% for ADNI.

• The paper presents a novel GCN framework that integrates imaging and non-imaging data to enhance disease prediction.
• It models populations as graphs where nodes represent individuals and edges capture phenotypic similarities.
• It achieves 70.4% accuracy for ASD and 80.0% for Alzheimer’s, outperforming conventional methods.

Disease Prediction using Graph Convolutional Networks: An Application to Autism Spectrum Disorder and Alzheimer's Disease

The paper under discussion presents a novel framework leveraging Graph Convolutional Networks (GCNs) for disease prediction, specifically targeting Autism Spectrum Disorder (ASD) and Alzheimer's disease (AD). The authors propose a methodology that integrates both imaging and non-imaging data into a graph-based model for large-scale population analysis, offering improvements over traditional approaches.

Overview of Methodology

The key innovation in this research is the utilization of GCNs to model populations as sparse graphs. In this configuration, each node represents an individual, characterized by imaging-based feature vectors, while edges encode phenotypic relationships using non-imaging data. This dual representation allows the framework to account for individual characteristics and similarities between subjects, which are essential in disease classification tasks.

Dataset and Evaluation

The framework was evaluated using two significant datasets: ABIDE (Autism Brain Imaging Data Exchange) and ADNI (Alzheimer's Disease Neuroimaging Initiative). ABIDE focuses on ASD classification using functional MRI and phenotypic data from 1112 subjects, whereas ADNI involves predicting the conversion from Mild Cognitive Impairment (MCI) to Alzheimer's disease through structural MRI data.

Experimental Findings

The results demonstrate a compelling performance, surpassing existing state-of-the-art methods. The framework achieves a classification accuracy of 70.4% for the ABIDE dataset and 80.0% for the ADNI dataset. These results underscore the importance of integrating graph-based approaches that respect both the imaging data and phenotypic context.

Implications and Future Work

The findings suggest that GCNs effectively capture complex interactions in medical datasets, improving prediction accuracy. This approach opens avenues for more precise diagnostic tools and could be extended to other types of disease prediction tasks. Future work may explore the integration of additional non-imaging features, optimizing graph construction methods, and exploring spectral methods such as Cayley polynomials for improved performance.

The implications for AI development are significant. The modeling of multimodal data in a graph structure could enhance AI's capabilities in handling diverse datasets, leading to broader applications in personalized medicine and large-scale population health studies. Moreover, the integration of domain-specific knowledge into graph structures emphasizes the potential for hybrid approaches that utilize both data-driven and expert-driven insights.

This research is a step forward in applying advanced computational models within the medical domain, offering a promising direction for enhancing diagnostic accuracy and understanding disease mechanisms through computational methods.