BPI-GNN: Interpretable brain network-based psychiatric diagnosis and subtyping.

Converging evidence increasingly suggests that psychiatric disorders, such as major depressive disorder (MDD) and autism spectrum disorder (ASD), are not unitary diseases, but rather heterogeneous syndromes that involve diverse, co-occurring symptoms and divergent responses to treatment. This clinical heterogeneity has hindered the progress of precision diagnosis and treatment effectiveness in psychiatric disorders. In this study, we propose BPI-GNN, a new interpretable graph neural network (GNN) framework for analyzing functional magnetic resonance images (fMRI), by leveraging the famed prototype learning. In addition, we introduce a novel generation process of prototype subgraph to discover essential edges of distinct prototypes and employ total correlation (TC) to ensure the independence of distinct prototype subgraph patterns. BPI-GNN can effectively discriminate psychiatric patients and healthy controls (HC), and identify biological meaningful subtypes of psychiatric disorders. We evaluate the performance of BPI-GNN against 11 popular brain network classification methods on three psychiatric datasets and observe that our BPI-GNN always achieves the highest diagnosis accuracy. More importantly, we examine differences in clinical symptom profiles and gene expression profiles among identified subtypes and observe that our identified brain-based subtypes have the clinical relevance. It also discovers the subtype biomarkers that align with current neuro-scientific knowledge.

CI-GNN: A Granger causality-inspired graph neural network for interpretable brain network-based psychiatric diagnosis.

There is a recent trend to leverage the power of graph neural networks (GNNs) for brain-network based psychiatric diagnosis, which, in turn, also motivates an urgent need for psychiatrists to fully understand the decision behavior of the used GNNs. However, most of the existing GNN explainers are either post-hoc in which another interpretive model needs to be created to explain a well-trained GNN, or do not consider the causal relationship between the extracted explanation and the decision, such that the explanation itself contains spurious correlations and suffers from weak faithfulness. In this work, we propose a granger causality-inspired graph neural network (CI-GNN), a built-in interpretable model that is able to identify the most influential subgraph (i.e., functional connectivity within brain regions) that is causally related to the decision (e.g., major depressive disorder patients or healthy controls), without the training of an auxillary interpretive network. CI-GNN learns disentangled subgraph-level representations α and ß that encode, respectively, the causal and non-causal aspects of original graph under a graph variational autoencoder framework, regularized by a conditional mutual information (CMI) constraint. We theoretically justify the validity of the CMI regulation in capturing the causal relationship. We also empirically evaluate the performance of CI-GNN against three baseline GNNs and four state-of-the-art GNN explainers on synthetic data and three large-scale brain disease datasets. We observe that CI-GNN achieves the best performance in a wide range of metrics and provides more reliable and concise explanations which have clinical evidence. The source code and implementation details of CI-GNN are freely available at GitHub repository (https://github.com/ZKZ-Brain/CI-GNN/).

Large-scale effective connectivity analysis reveals the existence of two mutual inhibitory systems in patients with major depression.

It is posited that cognitive and affective dysfunction in patients with major depression disorder (MDD) may be caused by dysfunctional signal propagation in the brain. By leveraging dynamic causal modeling, we investigated large-scale directed signal propagation (effective connectivity) among distributed large-scale brain networks with 43 MDD patients and 56 healthy controls. The results revealed the existence of two mutual inhibitory systems: the anterior default mode network, auditory network, sensorimotor network, salience network and visual networks formed an "emotional" brain, while the posterior default mode network, central executive networks, cerebellum and dorsal attention network formed a "rational brain". These two networks exhibited excitatory intra-system connectivity and inhibitory inter-system connectivity. Patients were characterized by potentiated intra-system connections within the "emotional/sensory brain", as well as over-inhibition of the "rational brain" by the "emotional/sensory brain". The hierarchical architecture of the large-scale effective connectivity networks was then analyzed using a PageRank algorithm which revealed a shift of the controlling role of the "rational brain" to the "emotional/sensory brain" in the patients. These findings inform basic organization of distributed large-scale brain networks and furnish a better characterization of the neural mechanisms of depression, which may facilitate effective treatment.

Aberrant temporal-spatial complexity of intrinsic fluctuations in major depression.

Accumulating evidence suggests that the brain is highly dynamic; thus, investigation of brain dynamics especially in brain connectivity would provide crucial information that stationary functional connectivity could miss. This study investigated temporal expressions of spatial modes within the default mode network (DMN), salience network (SN) and cognitive control network (CCN) using a reliable data-driven co-activation pattern (CAP) analysis in two independent data sets. We found enhanced CAP-to-CAP transitions of the SN in patients with MDD. Results suggested enhanced flexibility of this network in the patients. By contrast, we also found reduced spatial consistency and persistence of the DMN in the patients, indicating reduced variability and stability in individuals with MDD. In addition, the patients were characterized by prominent activation of mPFC. Moreover, further correlation analysis revealed that persistence and transitions of RCCN were associated with the severity of depression. Our findings suggest that functional connectivity in the patients may not be simply attenuated or potentiated, but just alternating faster or slower among more complex patterns. The aberrant temporal-spatial complexity of intrinsic fluctuations reflects functional diaschisis of resting-state networks as characteristic of patients with MDD.

Causal Interactions Between the Default Mode Network and Central Executive Network in Patients with Major Depression.

Two different but interacting neural systems exist in the human brain: the task positive networks and task negative networks. One of the most important task positive networks is the central executive network (CEN), while the task negative network generally refers to the default mode network (DMN), which usually demonstrates task-induced deactivation. Although previous studies have clearly shown the association of both the CEN and DMN with major depressive disorder (MDD), how the causal interactions between these two networks change in depressed patients remains unclear. In the current study, 99 subjects (43 patients with MDD and 56 healthy controls) were recruited with their resting-state fMRI data collected. After data preprocessing, spectral dynamic causal modeling (spDCM) was used to investigate the causal interactions within and between the DMN and CEN. Group commonalities and differences in causal interaction patterns within and between the CEN and DMN in patients and controls were assessed by a parametric empirical Bayes (PEB) model. Both subject groups demonstrated significant effective connectivity between regions of the CEN and DMN. In particular, we detected inhibitory influences from the CEN to the DMN with node-level PEB analyses, which may help to explain the anticorrelations between these two networks consistently reported in previous studies. Compared with healthy controls, patients with MDD showed increased effective connectivity within the CEN and decreased connectivity from regions of the CEN to DMN, suggesting impaired control of the DMN by the CEN in these patients. These findings might provide new insights into the neural substrates of MDD.

Quantitative Identification of Major Depression Based on Resting-State Dynamic Functional Connectivity: A Machine Learning Approach.

INTRODUCTION: Developing a machine learning-based approach which could provide quantitative identification of major depressive disorder (MDD) is essential for the diagnosis and intervention of this disorder. However, the performances of traditional algorithms using static functional connectivity (SFC) measures were unsatisfactory. In the present work, we exploit the hidden information embedded in dynamic functional connectivity (DFC) and developed an accurate and objective image-based diagnosis system for MDD. METHODS: MRI images were collected from 99 participants including 56 healthy controls and 43 MDD patients. DFC was calculated using a sliding-window algorithm. A non-linear support vector machine (SVM) approach was then used with the DFC matrices as features to distinguish MDD patients from healthy controls. The spatiotemporal characteristics of the most discriminative features were then investigated. RESULTS: The area under the curve (AUC) of the SVM classifier with DFC measures reached 0.9913, while this value is only 0.8685 for the algorithm using SFC measures. Spatially, the most discriminative 28 connections distributed in the visual network (VN), somatomotor network (SMN), dorsal attention network (DAN), ventral attention network (VAN), limbic network (LN), frontoparietal network (FPN), and default mode network (DMN), etc. Notably, a large portion of these connections were associated with the FPN, DMN, and VN. Temporally, the most discriminative connections transited from the cortex to deeper regions. CONCLUSION: The results clearly suggested that DFC is superior to SFC and provide a reliable quantitative identification method for MDD. Our findings may furnish a better understanding of the neural mechanisms of MDD as well as improve accurate diagnosis and early intervention of this disorder.

Structural networks analysis for depression combined with graph theory and the properties of fiber tracts via diffusion tensor imaging.

Previous studies have suggested that major depressive disorder was associated with topological properties of impaired white matter. However, most related studies only use one property of nerve fibers to construct whole-brain structural brain network. Considering white matter changes variously, We hypothesized whether the alternations of white matter topological properties could reflect different impairment of white matter integrity. In addition, it is still unknown whether impaired integrity of the white matter fiber tracts has relationship with abnormal topological properties in MDD. This study investigated the impaired white matter by using graph theoretic analyses in a cohort of 37 MDD patients and 38 matched control subjects. In addition, we further investigated fiber tracts differences in three interregional connectivity matrixes of significant different topological regions in MDD. Our graph theoretic analyses demonstrated that 7 different regions were observed for the local measures in patients with MDD compared with control groups. These regions were the central nodes of cortical-limbic network, frontal-cingulate network, default mode network (DMN), cognitive control network(CCN)and affective network (AN). In addition, two impaired white matter pathways which included inferior longitudinal fasciculus (ILF) and cingulum were observed in MDD using fiber tracts analysis. We speculate impaired integrity of ILF is due to the alternations in the number of axons or myelination. The results further demonstrated that the number of fiber tracts of anterior cingulum was associated with the depression scores in MDD.

Radiomics Strategy for Molecular Subtype Stratification of Lower-Grade Glioma: Detecting IDH and TP53 Mutations Based on Multimodal MRI.

BACKGROUND: Noninvasive detection of isocitrate dehydrogenase (IDH) and TP53 mutations are meaningful for molecular stratification of lower-grade gliomas (LrGG). PURPOSE: To explore potential MRI features reflecting IDH and TP53 mutations of LrGG, and propose a radiomics strategy for detecting them. STUDY TYPE: Retrospective, radiomics. POPULATION/SUBJECTS: A total of 103 LrGG patients were separated into development (n = 73) and validation (n = 30) cohorts. FIELD STRENGTH/SEQUENCE: T1 -weighted (before and after contrast-enhanced), T2 -weighted, and fluid-attenuation inversion recovery images from 1.5T (n = 37) or 3T (n = 66) scanners. ASSESSMENT: After data preprocessing, high-throughput features were derived from patients' volumes of interests of different sequences. The support vector machine-based recursive feature elimination (SVM-RFE) was adopted to find the optimal features for IDH and TP53 mutation detection. SVM models were trained and tested on development and validation cohort. The commonly used metric was used for assessing the efficiency. STATISTICAL TESTS: One-way analysis of variance (ANOVA), chi-square, or Fisher's exact test were applied on clinical characteristics to confirm whether significant differences exist between three molecular subtypes decided by IDH and TP53 status. Intraclass correlation coefficients were calculated to assess the robustness of the radiomics features. RESULTS: The constituent ratio of histopathologic subtypes was significantly different among three molecular subtypes (P = 0.017). SVM models for detecting IDH and TP53 mutation were established using 12 and 22 optimal features selected by SVM-RFE. The accuracies and area under the curves for IDH and TP53 mutations on the development cohort were 84.9%, 0.830, and 92.0%, 0.949, while on the validation cohort were 80.0%, 0.792, and 85.0%, 0.869, respectively. Furthermore, the stratified accuracies of three subtypes were 72.8%, 71.9%, and 70%, respectively. DATA CONCLUSION: Using a radiomics approach integrating SVM model and multimodal MRI features, molecular subtype stratification of LGG patients was implemented through detecting IDH and TP53 mutations. The results suggested that the proposed approach has promising detecting efficiency and T2 -weighted image features are more important than features from other images. LEVEL OF EVIDENCE: 3 Technical Efficacy: Stage 2 J. Magn. Reson. Imaging 2018;48:916-926.

Incapacity to control emotion in major depression may arise from disrupted white matter integrity and OFC-amygdala inhibition.

BACKGROUND: Disturbances in emotion regulation are the hallmarks of major depressive disorder (MDD). The incapacity to control negative emotion in patients has been associated with abnormal hyperactivation of the limbic system and hypoactivation of the frontal cortex. The amygdala and orbital frontal cortex (OFC) are two critical regions of the emotion regulation neural systems. METHODS: This study investigated the anatomical basis of abnormal emotion regulation by tracking the fiber tracts connecting the amygdala and OFC. In addition, using dynamic casual modeling on resting-state fMRI data of 20 MDD patients and equivalent controls, we investigated the exact neural mechanism through which abnormal communications between these two nodes were mediated in MDD. KEY RESULTS: The results revealed disrupted white matter integrity of fiber tracts in MDD, suggesting that functional abnormalities were accompanied by underlying anatomical basis. We also detected a failure of inhibition of the OFC on the activity of the amygdala in MDD, suggesting dysconnectivity was mediated through "top-down" influences from the frontal cortex to the amygdala. Following 8 weeks of antidepressant treatment, the patients showed significant clinical improvement and normalization of the abnormal OFC-amygdala structural and effective connectivity in the left hemisphere. CONCLUSIONS & INFERENCES: Our findings suggest that pathways connecting these two nodes may be core targets of the antidepressant treatment. In particular, it raised the intriguing question: Does the reversal of structural markers of connectivity reflect a response to antidepressant medication or activity-dependent myelination following a therapeutic restoration of effective connectivity?