Abstract:Recent applications of pattern recognition techniques on brain connectome classification using functional connectivity (FC) are shifting towards acknowledging the non-Euclidean topology and causal dynamics of brain connectivity across time. In this paper, a deep spatiotemporal variational Bayes (DSVB) framework is proposed to learn time-varying topological structures in dynamic FC networks for identifying autism spectrum disorder (ASD) in human participants. The framework incorporates a spatial-aware recurrent neural network with an attention-based message passing scheme to capture rich spatiotemporal patterns across dynamic FC networks. To overcome model overfitting on limited training datasets, an adversarial training strategy is introduced to learn graph embedding models that generalize well to unseen brain networks. Evaluation on the ABIDE resting-state functional magnetic resonance imaging dataset shows that our proposed framework substantially outperforms state-of-the-art methods in identifying patients with ASD. Dynamic FC analyses with DSVB-learned embeddings reveal apparent group differences between ASD and healthy controls in brain network connectivity patterns and switching dynamics of brain states.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to identify Autism Spectrum Disorder (ASD) through dynamic functional brain networks. Specifically, the paper proposes a Deep Spatiotemporal Variational Bayes (DSVB) framework, aiming to learn the time - varying topological structure in the Functional Connectivity (FC) network, so as to effectively distinguish autistic patients from healthy control groups. ### Background and Problems of the Paper 1. **Background**: - The human brain is a complex system composed of numerous interconnected neural regions. Functional magnetic resonance imaging (fMRI) shows that even in the resting state, there are spontaneous spatio - temporal fluctuations in brain activity. - Functional connectivity (FC) usually characterizes the blood - oxygen - level - dependent (BOLD) fMRI signals between brain regions at different spatial locations through statistical dependence (such as correlation). - Changes in the resting - state FC network are related to neuropsychiatric or neurodevelopmental disorders, such as Autism Spectrum Disorder (ASD). 2. **Limitations of Existing Methods**: - Traditional machine - learning methods (such as support vector machines, artificial neural networks, and deep - learning methods) usually model brain networks as two - dimensional grid - like connectivity matrices, ignoring the non - Euclidean nature of brain networks. - Graph neural networks (GNN) are mainly used for static brain - network modeling in these studies and lack methods for modeling dynamic aspects. ### The Method Proposed in the Paper The paper proposes a Deep Spatiotemporal Variational Bayes (DSVB) framework, with the following main contributions: 1. **End - to - End Probabilistic Framework**: Combine variational Bayes and Recurrent Graph Neural Networks (RNN) to learn the latent embedding space of serialized random graphs in a purely unsupervised manner, which can adapt to a wide range of spatio - temporal changes in unseen FC networks. 2. **Spatially - Aware Gated Recurrent Unit (GRU)**: Combine an attention - based message - passing scheme to generate hierarchical latent embeddings and extract the non - Euclidean topological structure and rich temporal patterns of the dynamic FC network. 3. **Adversarial Training Strategy**: Introduce an adversarial model perturbation strategy to regularize the GNN - based graph - embedding model and overcome the over - fitting problem caused by limited training data. 4. **Performance Verification**: Through nested - hierarchical 5 - fold cross - validation, it is proved that the proposed DSVB framework is superior to existing methods in identifying ASD from fMRI data. ### Experimental Results 1. **Performance Comparison**: - Table 1 shows the performance comparison between DSVB - FCNN and different static and dynamic baseline classifiers. The results show that DSVB - FCNN is significantly superior to other methods in multiple metrics such as accuracy, recall, precision, F1 - score, and AUC. - The standard deviation of DSVB - FCNN in all metrics is low, indicating that its performance has strong stability. 2. **Ablation Study**: - Table 2 shows the ablation study results of the DSVB framework. By removing certain components (such as variational Bayes, spatially - aware GRU, and adversarial training), the unique contributions of these components to improving the dynamic FC classification results are highlighted. - The results show that the spatially - aware GRU and adversarial training are crucial for improving model performance. ### Visualization Analysis - **t - SNE Visualization**: Figure 2 shows the t - SNE visualization results of the latent - state sequences and final readouts of different models, further verifying the importance of variational Bayes and adversarial training for learning a smooth latent - embedding space. - **Connection Pattern Analysis**: Figure 3 shows the average connection patterns derived from the embeddings learned by DSVB, revealing obvious differences between the healthy control group and the ASD group. In conclusion, through proposing the DSVB framework, this paper successfully solves the problem of identifying ASD from dynamic functional brain networks and achieves significant performance improvements in multiple aspects.

A Deep Probabilistic Spatiotemporal Framework for Dynamic Graph Representation Learning with Application to Brain Disorder Identification

Modeling the dynamic brain network representation for autism spectrum disorder diagnosis

Temporal Graph Representation Learning for Autism Spectrum Disorder Brain Networks

Multi-Scale Dynamic Graph Learning for Brain Disorder Detection with Functional MRI

DSAM: A Deep Learning Framework for Analyzing Temporal and Spatial Dynamics in Brain Networks

DynBrainGNN: Towards Spatio-Temporal Interpretable Graph Neural Network Based on Dynamic Brain Connectome for Psychiatric Diagnosis.

Reveal Consistent Spatial-Temporal Patterns from Dynamic Functional Connectivity for Autism Spectrum Disorder Identification.

Dynamic functional connections analysis with spectral learning for brain disorder detection

Towards Disease-Aware Self-Supervised Dynamic Brain Network Learning for Mental Diagnosis

Spatial-Temporal Characteristics of Brain Activity in Autism Spectrum Disorder Based on Hidden Markov Model and Dynamic Graph Theory: A Resting-State fMRI Study

A Spatial-Temporal Integration Analysis to Classify Dynamic Functional Connectivity for Brain Disease Diagnosis

Modeling dynamic characteristics of brain functional connectivity networks using resting-state functional MRI

An Improved Dynamic Functional Connectivity and Deep Neural Network Model for Autism Spectrum Disorder Classification

Deep Representations for Time-varying Brain Datasets

Topological and Graph Theoretical Analysis of Dynamic Functional Connectivity for Autism Spectrum Disorder

BrainTGL: A dynamic graph representation learning model for brain network analysis

Identifying Autism Spectrum Disorder Based on Individual-Aware Down-Sampling and Multi-Modal Learning

Identifying Autism Spectrum Disorder From Resting-State fMRI Using Deep Belief Network

A Unified Framework for Static and Dynamic Functional Connectivity Augmentation for Multi-Domain Brain Disorder Classification

MVS-GCN: A prior brain structure learning-guided multi-view graph convolution network for autism spectrum disorder diagnosis

Detection of autism spectrum disorder using graph representation learning algorithms and deep neural network, based on fMRI signals