TE-HI-GCN: An Ensemble of Transfer Hierarchical Graph Convolutional Networks for Disorder Diagnosis

A novel interactive deep cascade spectral graph convolutional network with multi-relational graphs for disease prediction

Graph neural networks (GNNs) have recently grown in popularity for disease prediction. Existing GNN-based methods primarily build the graph topological structure around a single modality and combine it with other modalities to acquire feature representations of acquisitions. The complicated relationship in each modality, however, may not be well highlighted due to its specificity. Further, relatively shallow networks restrict adequate extraction of high-level features, affecting disease prediction performance. Accordingly, this paper develops a new interactive deep cascade spectral graph convolutional network with multi-relational graphs (IDCGN) for disease prediction tasks. Its crucial points lie in constructing multiple relational graphs and dual cascade spectral graph convolution branches with interaction (DCSGBI). Specifically, the former designs a pairwise imaging-based edge generator and a pairwise non-imaging-based edge generator from different modalities by devising two learnable networks, which adaptively capture graph structures and provide various views of the same acquisition to aid in disease diagnosis. Again, DCSGBI is established to enrich high-level semantic information and low-level details of disease data. It devises a cascade spectral graph convolution operator for each branch and incorporates the interaction strategy between different branches into the network, successfully forming a deep model and capturing complementary information from diverse branches. In this manner, more favorable and sufficient features are learned for a reliable diagnosis. Experiments on several disease datasets reveal that IDCGN exceeds state-of-the-art models and achieves promising results.

Identifying Diagnostic Biomarkers for Autism Spectrum Disorder From Higher-order Interactions Using the PED Algorithm

In the field of neuroimaging, more studies of abnormalities in brain regions of the autism spectrum disorder (ASD) usually focused on two brain regions connected, and less on abnormalities of higher-order interactions of brain regions. To explore the complex relationships of brain regions, we used the partial entropy decomposition (PED) algorithm to capture higher-order interactions by computing the higher-order dependencies of all three brain regions (triads). We proposed a method for examining the effect of individual brain regions on triads based on the PED and surrogate tests. The key triads were discovered by analyzing the effects. Further, the hypergraph modularity maximization algorithm revealed the higher-order brain structures, of which the link between right thalamus and left thalamus in ASD was more loose compared with the typical control (TC). Redundant key triad (left cerebellum crus 1 and left precuneus and right inferior occipital gyrus) exhibited a discernible attenuation in interaction in ASD, while the synergistic key triad (right cerebellum crus 1 and left postcentral gyrus and left lingual gyrus) indicated a notable decline. The results of classification model further confirmed the potential of the key triads as diagnostic biomarkers.

The Use of Generative Adversarial Network and Graph Convolution Network for Neuroimaging-Based Diagnostic Classification

Functional connectivity (FC) obtained from resting-state functional magnetic resonance imaging has been integrated with machine learning algorithms to deliver consistent and reliable brain disease classification outcomes. However, in classical learning procedures, custom-built specialized feature selection techniques are typically used to filter out uninformative features from FC patterns to generalize efficiently on the datasets. The ability of convolutional neural networks (CNN) and other deep learning models to extract informative features from data with grid structure (such as images) has led to the surge in popularity of these techniques. However, the designs of many existing CNN models still fail to exploit the relationships between entities of graph-structure data (such as networks). Therefore, graph convolution network (GCN) has been suggested as a means for uncovering the intricate structure of brain network data, which has the potential to substantially improve classification accuracy. Furthermore, overfitting in classifiers can be largely attributed to the limited number of available training samples. Recently, the generative adversarial network (GAN) has been widely used in the medical field for its generative aspect that can generate synthesis images to cope with the problems of data scarcity and patient privacy. In our previous work, GCN and GAN have been designed to investigate FC patterns to perform diagnosis tasks, and their effectiveness has been tested on the ABIDE-I dataset. In this paper, the models will be further applied to FC data derived from more public datasets (ADHD, ABIDE-II, and ADNI) and our in-house dataset (PTSD) to justify their generalization on all types of data. The results of a number of experiments show the powerful characteristic of GAN to mimic FC data to achieve high performance in disease prediction. When employing GAN for data augmentation, the diagnostic accuracy across ADHD-200, ABIDE-II, and ADNI datasets surpasses that of other machine learning models, including results achieved with BrainNetCNN. Specifically, in ADHD, the accuracy increased from 67.74% to 73.96% with GAN, in ABIDE-II from 70.36% to 77.40%, and in ADNI, reaching 52.84% and 88.56% for multiclass and binary classification, respectively. GCN also obtains decent results, with the best accuracy in ADHD datasets at 71.38% for multinomial and 75% for binary classification, respectively, and the second-best accuracy in the ABIDE-II dataset (72.28% and 75.16%, respectively). Both GAN and GCN achieved the highest accuracy for the PTSD dataset, reaching 97.76%. However, there are still some limitations that can be improved. Both methods have many opportunities for the prediction and diagnosis of diseases.

Classification of Alzheimer's disease: application of a transfer learning deep Q-network method

Early diagnosis is crucial to slowing the progression of Alzheimer's disease (AD), so it is urgent to find an effective diagnostic method for AD. This study intended to investigate whether the transfer learning approach of deep Q-network (DQN) could effectively distinguish AD patients using local metrics of resting-state functional magnetic resonance imaging (rs-fMRI) as features. This study included 1310 subjects from the Consortium for Reliability and Reproducibility (CoRR) and 50 subjects from the Alzheimer's Disease Neuroimaging Initiative (ADNI) GO/2. The amplitude of low-frequency fluctuation (ALFF), fractional ALFF (fALFF) and percent amplitude of fluctuation (PerAF) were extracted as features using the Power 264 atlas. Based on gender bias in AD, we searched for transferable similar parts between the CoRR feature matrix and the ADNI feature matrix, resulting in the CoRR similar feature matrix served as the source domain and the ADNI similar feature matrix served as the target domain. A DQN classifier was pre-trained in the source domain and transferred to the target domain. Finally, the transferred DQN classifier was used to classify AD and healthy controls (HC). A permutation test was performed. The DQN transfer learning achieved a classification accuracy of 86.66% (p < 0.01), recall of 83.33% and precision of 83.33%. The findings suggested that the transfer learning approach using DQN could be an effective way to distinguish AD from HC. It also revealed the potential value of local brain activity in AD clinical diagnosis.

Mapping Multi-Modal Brain Connectome for Brain Disorder Diagnosis via Cross-Modal Mutual Learning

Recently, the study of multi-modal brain connectome has recorded a tremendous increase and facilitated the diagnosis of brain disorders. In this paradigm, functional and structural networks, e.g., functional and structural connectivity derived from fMRI and DTI, are in some manner interacted but are not necessarily linearly related. Accordingly, there remains a great challenge to leverage complementary information for brain connectome analysis. Recently, Graph Convolutional Networks (GNN) have been widely applied to the fusion of multi-modal brain connectome. However, most existing GNN methods fail to couple inter-modal relationships. In this regard, we propose a Cross-modal Graph Neural Network (Cross-GNN) that captures inter-modal dependencies through dynamic graph learning and mutual learning. Specifically, the inter-modal representations are attentively coupled into a compositional space for reasoning inter-modal dependencies. Additionally, we investigate mutual learning in explicit and implicit ways: (1) Cross-modal representations are obtained by cross-embedding explicitly based on the inter-modal correspondence matrix. (2) We propose a cross-modal distillation method to implicitly regularize latent representations with cross-modal semantic contexts. We carry out statistical analysis on the attentively learned correspondence matrices to evaluate inter-modal relationships for associating disease biomarkers. Our extensive experiments on three datasets demonstrate the superiority of our proposed method for disease diagnosis with promising prediction performance and multi-modal connectome biomarker location.

TSP-GNN: a novel neuropsychiatric disorder classification framework based on task-specific prior knowledge and graph neural network

Neuropsychiatric disorder (ND) is often accompanied by abnormal functional connectivity (FC) patterns in specific task contexts. The distinctive task-specific FC patterns can provide valuable features for ND classification models using deep learning. However, most previous studies rely solely on the whole-brain FC matrix without considering the prior knowledge of task-specific FC patterns. Insight by the decoding studies on brain-behavior relationship, we develop TSP-GNN, which extracts task-specific prior (TSP) connectome patterns and employs graph neural network (GNN) for disease classification. TSP-GNN was validated using publicly available datasets. Our results demonstrate that different ND types show distinct task-specific connectivity patterns. Compared with the whole-brain node characteristics, utilizing task-specific nodes enhances the accuracy of ND classification. TSP-GNN comprises the first attempt to incorporate prior task-specific connectome patterns and the power of deep learning. This study elucidates the association between brain dysfunction and specific cognitive processes, offering valuable insights into the cognitive mechanism of neuropsychiatric disease.

Do it the transformer way: A comprehensive review of brain and vision transformers for autism spectrum disorder diagnosis and classification

Autism spectrum disorder (ASD) is a condition observed in children who display abnormal patterns of interaction, behavior, and communication with others. Despite extensive research efforts, the underlying causes of this neurodevelopmental disorder and its biomarkers remain unknown. However, advancements in artificial intelligence and machine learning have improved clinicians' ability to diagnose ASD. This review paper investigates various MRI modalities to identify distinct features that characterize individuals with ASD compared to typical control subjects. The review then moves on to explore deep learning models for ASD diagnosis, including convolutional neural networks (CNNs), autoencoders, graph convolutions, attention networks, and other models. CNNs and their variations are particularly effective due to their capacity to learn structured image representations and identify reliable biomarkers for brain disorders. Computer vision transformers often employ CNN architectures with transfer learning techniques like fine-tuning and layer freezing to enhance image classification performance, surpassing traditional machine learning models. This review paper contributes in three main ways. Firstly, it provides a comprehensive overview of a recommended architecture for using vision transformers in the systematic ASD diagnostic process. To this end, the paper investigates various pre-trained vision architectures such as VGG, ResNet, Inception, InceptionResNet, DenseNet, and Swin models that were fine-tuned for ASD diagnosis and classification. Secondly, it discusses the vision transformers of 2020th like BiT, ViT, MobileViT, and ConvNeXt, and applying transfer learning methods in relation to their prospective practicality in ASD classification. Thirdly, it explores brain transformers that are pre-trained on medically rich data and MRI neuroimaging datasets. The paper recommends a systematic architecture for ASD diagnosis using brain transformers. It also reviews recently developed brain transformer-based models, such as METAFormer, Com-BrainTF, Brain Network, ST-Transformer, STCAL, BolT, and BrainFormer, discussing their deep transfer learning architectures and results in ASD detection. Additionally, the paper summarizes and discusses brain-related transformers for various brain disorders, such as MSGTN, STAGIN, and MedTransformer, in relation to their potential usefulness in ASD. The study suggests that developing specialized transformer-based models, following the success of natural language processing (NLP), can offer new directions for image classification problems in ASD brain biomarkers learning and classification. By incorporating the attention mechanism, treating MRI modalities as sequence prediction tasks trained on brain disorder classification problems, and fine-tuned on ASD datasets, brain transformers can show a great promise in ASD diagnosis.

The Combination of a Graph Neural Network Technique and Brain Imaging to Diagnose Neurological Disorders: A Review and Outlook

Neurological disorders (NDs), such as Alzheimer's disease, have been a threat to human health all over the world. It is of great importance to diagnose ND through combining artificial intelligence technology and brain imaging. A graph neural network (GNN) can model and analyze the brain, imaging from morphology, anatomical structure, function features, and other aspects, thus becoming one of the best deep learning models in the diagnosis of ND. Some researchers have investigated the application of GNN in the medical field, but the scope is broad, and its application to NDs is less frequent and not detailed enough. This review focuses on the research progress of GNNs in the diagnosis of ND. Firstly, we systematically investigated the GNN framework of ND, including graph construction, graph convolution, graph pooling, and graph prediction. Secondly, we investigated common NDs using the GNN diagnostic model in terms of data modality, number of subjects, and diagnostic accuracy. Thirdly, we discussed some research challenges and future research directions. The results of this review may be a valuable contribution to the ongoing intersection of artificial intelligence technology and brain imaging.

A deep connectome learning network using graph convolution for connectome-disease association study

Multivariate analysis approaches provide insights into the identification of phenotype associations in brain connectome data. In recent years, deep learning methods including convolutional neural network (CNN) and graph neural network (GNN), have shifted the development of connectome-wide association studies (CWAS) and made breakthroughs for connectome representation learning by leveraging deep embedded features. However, most existing studies remain limited by potentially ignoring the exploration of region-specific features, which play a key role in distinguishing brain disorders with high intra-class variations, such as autism spectrum disorder (ASD), and attention deficit hyperactivity disorder (ADHD). Here, we propose a multivariate distance-based connectome network (MDCN) that addresses the local specificity problem by efficient parcellation-wise learning, as well as associating population and parcellation dependencies to map individual differences. The approach incorporating an explainable method, parcellation-wise gradient and class activation map (p-GradCAM), is feasible for identifying individual patterns of interest and pinpointing connectome associations with diseases. We demonstrate the utility of our method on two largely aggregated multicenter public datasets by distinguishing ASD and ADHD from healthy controls and assessing their associations with underlying diseases. Extensive experiments have demonstrated the superiority of MDCN in classification and interpretation, where MDCN outperformed competitive state-of-the-art methods and achieved a high proportion of overlap with previous findings. As a CWAS-guided deep learning method, our proposed MDCN framework may narrow the bridge between deep learning and CWAS approaches, and provide new insights for connectome-wide association studies.

Discriminative analysis of schizophrenia patients using graph convolutional networks: A combined multimodal MRI and connectomics analysis

INTRODUCTION

Recent studies in human brain connectomics with multimodal magnetic resonance imaging (MRI) data have widely reported abnormalities in brain structure, function and connectivity associated with schizophrenia (SZ). However, most previous discriminative studies of SZ patients were based on MRI features of brain regions, ignoring the complex relationships within brain networks.

METHODS

We applied a graph convolutional network (GCN) to discriminating SZ patients using the features of brain region and connectivity derived from a combined multimodal MRI and connectomics analysis. Structural magnetic resonance imaging (sMRI) and resting-state functional magnetic resonance imaging (rs-fMRI) data were acquired from 140 SZ patients and 205 normal controls. Eighteen types of brain graphs were constructed for each subject using 3 types of node features, 3 types of edge features, and 2 brain atlases. We investigated the performance of 18 brain graphs and used the TopK pooling layers to highlight salient brain regions (nodes in the graph).

RESULTS

The GCN model, which used functional connectivity as edge features and multimodal features (sMRI + fMRI) of brain regions as node features, obtained the highest average accuracy of 95.8%, and outperformed other existing classification studies in SZ patients. In the explainability analysis, we reported that the top 10 salient brain regions, predominantly distributed in the prefrontal and occipital cortices, were mainly involved in the systems of emotion and visual processing.

DISCUSSION

Our findings demonstrated that GCN with a combined multimodal MRI and connectomics analysis can effectively improve the classification of SZ at an individual level, indicating a promising direction for the diagnosis of SZ patients. The code is available at https://github.com/CXY-scut/GCN-SZ.git.

Evaluation of Risk of Bias in Neuroimaging-Based Artificial Intelligence Models for Psychiatric Diagnosis: A Systematic Review

IMPORTANCE

Neuroimaging-based artificial intelligence (AI) diagnostic models have proliferated in psychiatry. However, their clinical applicability and reporting quality (ie, feasibility) for clinical practice have not been systematically evaluated.

OBJECTIVE

To systematically assess the risk of bias (ROB) and reporting quality of neuroimaging-based AI models for psychiatric diagnosis.

EVIDENCE REVIEW

PubMed was searched for peer-reviewed, full-length articles published between January 1, 1990, and March 16, 2022. Studies aimed at developing or validating neuroimaging-based AI models for clinical diagnosis of psychiatric disorders were included. Reference lists were further searched for suitable original studies. Data extraction followed the CHARMS (Checklist for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modeling Studies) and PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-analyses) guidelines. A closed-loop cross-sequential design was used for quality control. The PROBAST (Prediction Model Risk of Bias Assessment Tool) and modified CLEAR (Checklist for Evaluation of Image-Based Artificial Intelligence Reports) benchmarks were used to systematically evaluate ROB and reporting quality.

FINDINGS

A total of 517 studies presenting 555 AI models were included and evaluated. Of these models, 461 (83.1%; 95% CI, 80.0%-86.2%) were rated as having a high overall ROB based on the PROBAST. The ROB was particular high in the analysis domain, including inadequate sample size (398 of 555 models [71.7%; 95% CI, 68.0%-75.6%]), poor model performance examination (with 100% of models lacking calibration examination), and lack of handling data complexity (550 of 555 models [99.1%; 95% CI, 98.3%-99.9%]). None of the AI models was perceived to be applicable to clinical practices. Overall reporting completeness (ie, number of reported items/number of total items) for the AI models was 61.2% (95% CI, 60.6%-61.8%), and the completeness was poorest for the technical assessment domain with 39.9% (95% CI, 38.8%-41.1%).

CONCLUSIONS AND RELEVANCE

This systematic review found that the clinical applicability and feasibility of neuroimaging-based AI models for psychiatric diagnosis were challenged by a high ROB and poor reporting quality. Particularly in the analysis domain, ROB in AI diagnostic models should be addressed before clinical application.

A Convolutional Neural Network and Graph Convolutional Network Based Framework for AD Classification

The neuroscience community has developed many convolutional neural networks (CNNs) for the early detection of Alzheimer's disease (AD). Population graphs are thought of as non-linear structures that capture the relationships between individual subjects represented as nodes, which allows for the simultaneous integration of imaging and non-imaging information as well as individual subjects' features. Graph convolutional networks (GCNs) generalize convolution operations to accommodate non-Euclidean data and aid in the mining of topological information from the population graph for a disease classification task. However, few studies have examined how GCNs' input properties affect AD-staging performance. Therefore, we conducted three experiments in this work. Experiment 1 examined how the inclusion of demographic information in the edge-assigning function affects the classification of AD versus cognitive normal (CN). Experiment 2 was designed to examine the effects of adding various neuropsychological tests to the edge-assigning function on the mild cognitive impairment (MCI) classification. Experiment 3 studied the impact of the edge assignment function. The best result was obtained in Experiment 2 on multi-class classification (AD, MCI, and CN). We applied a novel framework for the diagnosis of AD that integrated CNNs and GCNs into a unified network, taking advantage of the excellent feature extraction capabilities of CNNs and population-graph processing capabilities of GCNs. To learn high-level anatomical features, DenseNet was used; a set of population graphs was represented with nodes defined by imaging features and edge weights determined by different combinations of imaging or/and non-imaging information, and the generated graphs were then fed to the GCNs for classification. Both binary classification and multi-class classification showed improved performance, with an accuracy of 91.6% for AD versus CN, 91.2% for AD versus MCI, 96.8% for MCI versus CN, and 89.4% for multi-class classification. The population graph's imaging features and edge-assigning functions can both significantly affect classification accuracy.

Classification of Brain Disorders in rs-fMRI via Local-to-Global Graph Neural Networks

Recently, functional brain network has been used for the classification of brain disorders, such as Autism Spectrum Disorder (ASD) and Alzheimer's disease (AD). Existing methods either ignore the non-imaging information associated with the subjects and the relationship between the subjects, or cannot identify and analyze disease-related local brain regions and biomarkers, leading to inaccurate classification results. This paper proposes a local-to-global graph neural network (LG-GNN) to address this issue. A local ROI-GNN is designed to learn feature embeddings of local brain regions and identify biomarkers, and a global Subject-GNN is then established to learn the relationship between the subjects with the embeddings generated by the local ROI-GNN and the non-imaging information. The local ROI-GNN contains a self-attention based pooling module to preserve the embeddings most important for the classification. The global Subject-GNN contains an adaptive weight aggregation block to generate the multi-scale feature embedding corresponding to each subject. The proposed LG-GNN is thoroughly validated using two public datasets for ASD and AD classification. The experimental results demonstrated that it achieves the state-of-the-art performance in terms of various evaluation metrics.

Multi-View Feature Enhancement Based on Self-Attention Mechanism Graph Convolutional Network for Autism Spectrum Disorder Diagnosis

Functional connectivity (FC) network based on resting-state functional magnetic resonance imaging (rs-fMRI) has become an important tool to explore and understand the brain, which can provide objective basis for the diagnosis of neurodegenerative diseases, such as autism spectrum disorder (ASD). However, most functional connectivity (FC) networks only consider the unilateral features of nodes or edges, and the interaction between them is ignored. In fact, their integration can provide more comprehensive and crucial information in the diagnosis. To address this issue, a new multi-view brain network feature enhancement method based on self-attention mechanism graph convolutional network (SA-GCN) is proposed in this article, which can enhance node features through the connection relationship among different nodes, and then extract deep-seated and more discriminative features. Specifically, we first plug the pooling operation of self-attention mechanism into graph convolutional network (GCN), which can consider the node features and topology of graph network at the same time and then capture more discriminative features. In addition, the sample size is augmented by a "sliding window" strategy, which is beneficial to avoid overfitting and enhance the generalization ability. Furthermore, to fully explore the complex connection relationship among brain regions, we constructed the low-order functional graph network (Lo-FGN) and the high-order functional graph network (Ho-FGN) and enhance the features of the two functional graph networks (FGNs) based on SA-GCN. The experimental results on benchmark datasets show that: (1) SA-GCN can play a role in feature enhancement and can effectively extract more discriminative features, and (2) the integration of Lo-FGN and Ho-FGN can achieve the best ASD classification accuracy (79.9%), which reveals the information complementarity between them.

Transfer Learning Approaches for Neuroimaging Analysis: A Scoping Review

Deep learning algorithms have been moderately successful in diagnoses of diseases by analyzing medical images especially through neuroimaging that is rich in annotated data. Transfer learning methods have demonstrated strong performance in tackling annotated data. It utilizes and transfers knowledge learned from a source domain to target domain even when the dataset is small. There are multiple approaches to transfer learning that result in a range of performance estimates in diagnosis, detection, and classification of clinical problems. Therefore, in this paper, we reviewed transfer learning approaches, their design attributes, and their applications to neuroimaging problems. We reviewed two main literature databases and included the most relevant studies using predefined inclusion criteria. Among 50 reviewed studies, more than half of them are on transfer learning for Alzheimer's disease. Brain mapping and brain tumor detection were second and third most discussed research problems, respectively. The most common source dataset for transfer learning was ImageNet, which is not a neuroimaging dataset. This suggests that the majority of studies preferred pre-trained models instead of training their own model on a neuroimaging dataset. Although, about one third of studies designed their own architecture, most studies used existing Convolutional Neural Network architectures. Magnetic Resonance Imaging was the most common imaging modality. In almost all studies, transfer learning contributed to better performance in diagnosis, classification, segmentation of different neuroimaging diseases and problems, than methods without transfer learning. Among different transfer learning approaches, fine-tuning all convolutional and fully-connected layers approach and freezing convolutional layers and fine-tuning fully-connected layers approach demonstrated superior performance in terms of accuracy. These recent transfer learning approaches not only show great performance but also require less computational resources and time.