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.

Deep Neural Networks and Kernel Regression Achieve Comparable Accuracies for Functional Connectivity Prediction of Behavior and Demographics

“…However, we note that all algorithms performed worse than simply predicting the median pairs matching value in the training set, which would have yielded an MAE of 0.4. (Kawahara et al, 2017;Parisot et al, 2017Parisot et al, , 2018 versus our tuned hyperparameters (Figure 4). Results using our hyperparameters compared favorably with the results using the reference hyperparameters.…”

“…Results using our hyperparameters compared favorably with the results using the reference hyperparameters. (Kawahara et al, 2017;Parisot et al, 2017Parisot et al, , 2018 versus our tuned hyperparameters (Figure 7). Results using our hyperparameters compared favorably with the results using the reference hyperparameters (except for GCNN and pairs matching).…”

Improving multi-site autism classification based on site-dependence minimisation and second-order functional connectivity

“…We included rs-fMRI and phenotypic data from 505 ASD and 530 TC individuals, yielding a sample 110 of 1035 subjects. This sample of subjects is the same as that used in (Heinsfeld et al, 2018), which differs from the 871 subjects used in (Parisot et al, 2018;Abraham et al, 2017) due to their use of image quality control measures upon Subject brain fMRI data features. An illustration of all possible models studied in this paper.…”

“…25 While it presents a great potential for the extraction of functional biomarkers for autism classification, its multi-site and multi-protocol aspects bring along 2 significant patient heterogeneity, statistical noise and experimental differences in the rs-fMRI data, making the classification task much more challenging (Abraham et al (2017). Recent works have employed different ML methods, such as 30 recurrent neural networks (RNN), graph convolutional neural networks (GCN) and denoising autoencoders (Dvornek et al, 2018;Heinsfeld et al, 2018;Ktena et al, 2018;Parisot et al, 2018). However, despite the complexity in patterns that these methods can generally capture, the difference in their top classification results on ABIDE fall less than 1%, with the highest achieved accuracy 35 being 70.4% by the GCN model developed by Parisot et al (2018).…”

“…Recent works have employed different ML methods, such as 30 recurrent neural networks (RNN), graph convolutional neural networks (GCN) and denoising autoencoders (Dvornek et al, 2018;Heinsfeld et al, 2018;Ktena et al, 2018;Parisot et al, 2018). However, despite the complexity in patterns that these methods can generally capture, the difference in their top classification results on ABIDE fall less than 1%, with the highest achieved accuracy 35 being 70.4% by the GCN model developed by Parisot et al (2018). This paper investigates two research questions that can potentially improve multi-site autism classification.…”