skip to main content

Title: Learning Multi-resolution Graph Edge Embedding for Discovering Brain Network Dysfunction in Neurological Disorders
Tremendous recent literature show that associations between different brain regions, i.e., brain connectivity, provide early symptoms of neurological disorders. Despite significant efforts made for graph neural network (GNN) techniques, their focus on graph nodes makes the state-of-the-art GNN methods not suitable for classifying brain connectivity as graphs where the objective is to characterize disease-relevant network dysfunction patterns on graph links. To address this issue, we propose Multi-resolution Edge Network (MENET) to detect disease-specific connectomic benchmarks with high discrimination power across diagnostic categories. The core of MENET is a novel graph edge-wise transform that we propose, which allows us to capture multi-resolution “connectomic” features. Using a rich set of the connectomic features, we devise a graph learning framework to jointly select discriminative edges and assign diagnostic labels for graphs. Experiments on two real datasets show that MENET accurately predicts diagnostic labels and identify brain connectivities highly associated with neurological disorders such as Alzheimer’s Disease and Attention-Deficit/Hyperactivity Disorder.  more » « less
Award ID(s):
1948510 2008602
Author(s) / Creator(s):
; ; ;
Date Published:
Journal Name:
Lecture notes in computer science
Page Range / eLocation ID:
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Graph Neural Networks (GNN) offer the powerful approach to node classification in complex networks across many domains including social media, E-commerce, and FinTech. However, recent studies show that GNNs are vulnerable to attacks aimed at adversely impacting their node classification performance. Existing studies of adversarial attacks on GNN focus primarily on manipulating the connectivity between existing nodes, a task that requires greater effort on the part of the attacker in real-world applications. In contrast, it is much more expedient on the part of the attacker to inject adversarial nodes, e.g., fake profiles with forged links, into existing graphs so as to reduce the performance of the GNN in classifying existing nodes. Hence, we consider a novel form of node injection poisoning attacks on graph data. We model the key steps of a node injection attack, e.g., establishing links between the injected adversarial nodes and other nodes, choosing the label of an injected node, etc. by a Markov Decision Process. We propose a novel reinforcement learning method for Node Injection Poisoning Attacks (NIPA), to sequentially modify the labels and links of the injected nodes, without changing the connectivity between existing nodes. Specifically, we introduce a hierarchical Q-learning network to manipulate the labels of the adversarial nodes and their links with other nodes in the graph, and design an appropriate reward function to guide the reinforcement learning agent to reduce the node classification performance of GNN. The results of the experiments show that NIPA is consistently more effective than the baseline node injection attack methods for poisoning graph data on three benchmark datasets. 
    more » « less
  2. INTRODUCTION The analysis of the human brain is a central goal of neuroscience, but for methodological reasons, research has focused on model organisms, the mouse in particular. Because substantial homology was found at the level of ion channels, transcriptional programs, and basic neuronal types, a strong similarity of neuronal circuits across species has also been assumed. However, a rigorous test of the configuration of local neuronal circuitry in mouse versus human—in particular, in the gray matter of the cerebral cortex—is missing. The about 1000-fold increase in number of neurons is the most obvious evolutionary change of neuronal network properties from mouse to human. Whether the structure of the local cortical circuitry has changed as well is, however, unclear. Recent data from transcriptomic analyses has indicated an increase in the proportion of inhibitory interneurons from mouse to human. But what the effect of such a change is on the circuit configurations found in the human cerebral cortex is not known. This is, however, of particular interest also to the study of neuropsychiatric disorders because in these, the alteration of inhibitory-to-excitatory synaptic balance has been identified as one possible mechanistic underpinning. RATIONALE We used recent methodological improvements in connectomics to acquire data from one macaque and two human individuals, using biopsies of the temporal, parietal, and frontal cortex. Human tissue was obtained from neurosurgical interventions related to tumor removal, in which access path tissue was harvested that was not primarily affected by the underlying disease. A key concern in the analysis of human patient tissue has been the relation to epilepsy surgery, when the underlying disease has required often year-long treatment with pharmaceuticals, plausibly altering synaptic connectivity. Therefore, the analysis of nonepileptic surgery tissue seemed of particular importance. We also included data from one macaque individual, who was not known to have any brain-related pathology. RESULTS We acquired three-dimensional electron microscopy data from temporal and frontal cortex of human and temporal and parietal cortex of macaque. From these, we obtained connectomic reconstructions and compared these with five connectomes from mouse cortex. On the basis of these data, we were able to determine the effect of the about 2.5-fold expansion of the interneuron pool in macaque and human cortex compared with that of mouse. Contrary to expectation, the inhibitory-to-excitatory synaptic balance on pyramidal neurons in macaque and human cortex was not substantially altered. Rather, the interneuron pool was selectively expanded for bipolar-type interneurons, which prefer the innervation of other interneurons, and which further increased their preference for interneuron innervation from mouse to human. These changes were each multifold, yielding in effect an about 10-fold expanded interneuron-to-interneuron network in the human cortex that is only sparsely present in mouse. The total amount of synaptic input to pyramidal neurons, however, did not change according to the threefold thickening of the cortex; rather, a modest increase from about 12,000 synaptic inputs in mouse to about 15,000 in human was found. CONCLUSION The principal cells of the cerebral cortex, pyramidal neurons, maintain almost constant inhibitory-to-excitatory input balance and total synaptic input across 100 million years of evolutionary divergence, which is particularly noteworthy with the concomitant 1000-fold expansion of the neuronal network size and the 2.5-fold increase of inhibitory interneurons from mouse to human. Rather, the key network change from mouse to human is an expansion of almost an order of magnitude of an interneuron-to-interneuron network that is virtually absent in mouse but constitutes a substantial part of the human cortical network. Whether this new network is primarily created through the expansion of existing neuronal types, or is related to the creation of new interneuron subtypes, requires further study. The discovery of this network component in human cortex encourages detailed analysis of its function in health and disease. Connectomic screening across mammalian species: Comparison of five mouse, two macaque, and two human connectomic datasets from the cerebral cortex. ( A ) Automated reconstructions of all neurons with their cell bodies in the volume shown, using random colors. The analyzed connectomes comprised a total of ~1.6 million synapses. Arrows indicate evolutionary divergence: the last common ancestor between human and mouse, approximately 100 million years ago, and the last common ancestor between human and macaque, about 20 million years ago. ( B ) Illustration of the about 10-fold expansion of the interneuron-to-interneuron network from mouse to human. 
    more » « less
  3. Brain functional network connectivity is an important measure for characterizing changes in a variety of neurological disorders, for example Alzheimer’s Disease, Parkinson Disease, and Epilepsy. Epilepsy is a serious neurological disorder affecting more than 50 million persons worldwide with severe impact on the quality of life of patients and their family members due to recurrent seizures. More than 30% of epilepsy patients are refractive to pharmacotherapy and are considered for resection to disrupt epilepsy seizure networks. However, 20-50% of these patients continue to have seizures after surgery. Therefore, there is a critical need to gain new insights into the characteristics of epilepsy seizure networks involving one of more brain regions and accurately delineate epileptogenic zone as a target for surgery. Although there is growing availability of large volume of high resolution stereotactic electroencephalogram (SEEG) data recorded from intracranial electrodes during presurgical evaluation of patients, there are significant informatics challenges associated with processing and analyzing this large signal dataset for characterizing epilepsy seizure networks. In this paper, we describe the development and application of a high-performance indexing structure for efficient retrieval of large-scale SEEG signal data to compute seizure network patterns corresponding to brain functional connectivity networks. This novel Neuro-Integrative Connectivity (NIC) search and retrieval method has been developed by extending the red-black tree index model together with an efficient lookup algorithm. We systematically perform a comparative evaluation of the proposed NIC index using de-identified SEEG data from a patient with temporal lobe epilepsy to retrieve segments of signal data corresponding to multiple seizure events and demonstrate the significant advantages of the NIC index as compared to existing methods. This new NIC Index enables faster computation of brain functional connectivity measures in epilepsy patients for large-scale network analysis and potentially provide new insights into the organization as well as evolution of seizure networks in epilepsy patients. 
    more » « less
  4. null (Ed.)
    In this paper, we propose a supervised graph representation learning method to model the relationship between brain functional connectivity (FC) and structural connectivity (SC) through a graph encoder-decoder system. The graph convolutional network (GCN) model is leveraged in the encoder to learn lower-dimensional node representations (i.e. node embeddings) integrating information from both node attributes and network topology. In doing so, the encoder manages to capture both direct and indirect interactions between brain regions in the node embeddings which later help reconstruct empirical FC networks. From node embeddings, graph representations are learnt to embed the entire graphs into a vector space. Our end-to-end model utilizes a multi-objective loss function to simultaneously learn node representations for FC network reconstruction and graph representations for subject classification. The experiment on a large population of non-drinkers and heavy drinkers shows that our model can provide a characterization of the population pattern in the SC-FC relationship, while also learning features that capture individual uniqueness for subject classification. The identified key brain subnetworks show significant between-group difference and support the promising prospect of GCN-based graph representation learning on brain networks to model human brain activity and function. 
    more » « less
  5. Deep learning has been applied to magnetic resonance imaging (MRI) for a variety of purposes, ranging from the acceleration of image acquisition and image denoising to tissue segmentation and disease diagnosis. Convolutional neural networks have been particularly useful for analyzing MRI data due to the regularly sampled spatial and temporal nature of the data. However, advances in the field of brain imaging have led to network- and surface-based analyses that are often better represented in the graph domain. In this analysis, we propose a general purpose cortical segmentation method that, given resting-state connectivity features readily computed during conventional MRI pre-processing and a set of corresponding training labels, can generate cortical parcellations for new MRI data. We applied recent advances in the field of graph neural networks to the problem of cortical surface segmentation, using resting-state connectivity to learn discrete maps of the human neocortex. We found that graph neural networks accurately learn low-dimensional representations of functional brain connectivity that can be naturally extended to map the cortices of new datasets. After optimizing over algorithm type, network architecture, and training features, our approach yielded mean classification accuracies of 79.91% relative to a previously published parcellation. We describe how some hyperparameter choices including training and testing data duration, network architecture, and algorithm choice affect model performance. 
    more » « less