To read the full PDF, click here
Anthony King, PhD
Yana Lokshina, MS
Chandra Sripada, MD, PhD
Israel Liberzon, MD
Anthony King, PhD
Assistant Research Professor
Department of Psychiatry
University of Michigan
Ann Arbor, MI 48105
Posttraumatic stress disorder (PTSD) is common and is associated with suffering, disability, and huge human and economic costs to society. Neuroimaging studies over the past 20 years have pointed to potential dysfunction in several brain regions, including amygdala, insula, and frontal cortex, however they have been limited by relatively small N. Furthermore, accumulating evidence converges on the view that neural underpinnings of psychiatric disorders like PTSD may be reflected in dysfunctions in relationships between large-scale, distributed neural networks. Recent advances in resting state functional connectivity (rsFC) methodology have allowed for testing effects psychiatric disorders on large-scale networks, using both a priori “seed” based approaches, as well as unbiased whole brain “connectome-wide” approaches. Evidence from several seed-based rsFC studies suggests PTSD is associated with increased functional connectivity within the Salience Network (associated with detection of salience and threat) and decreased functional connectivity within the Default Mode Network (associated with social cognition and autobiographical memory). Furthermore, PTSD shows aberrantly increased connectivity or “desegregation” between these two networks, which are usually anti-correlated in healthy people. Emerging data from recent a priori seed-based and connectome-wide studies also suggest potential contribution of changes in other networks, including attention and cognitive control networks in PTSD. We propose to use the ENIGMA-PTSD sample to conduct the largest confirmatory “mega-analysis” to test a priori hypotheses about the role of altered SN and DMN connectivity in PTSD, and to also conduct exploratory analyses using unbiased whole-brain connectomic methodologies to identify potential contributions of additional networks, and elucidate the effects of PTSD on global neural network architecture.
Background: Over 20 years of relatively small neuroimaging studies (SPECT, PET, fMRI) have revealed what appear to be reliable and replicable alterations in several brain regions, including, insula, dmPFC, dorsal ACC, and hippocampus (for reviews and meta-analyses see1-3). Mounting evidence from basic, translational, and treatment research converges on the view that neural underpinnings of psychiatric disorders like PTSD may be better understood by dysfunctions in relationships between large-scale neural networks4-7. The ENIGMA PGC-PTSD Neuroimaging group offers an unprecedented opportunity to (1) confirm these initial findings and (2) to conduct large N exploratory, well powered studies to identify additional potential targets.
Summary: Thus, our approach will be to conduct (1) confirmatory tests of a priori hypotheses regarding the role of desegregation of the Default Mode Network (DMN) and Salience Network (SN), and changes in connectivity within these networks, in PTSD (based upon prior evidence, in the literature described below) using seed-based approaches, and (2) exploratory analyses using connectome-wide approaches to identify potential dysfunction in other networks (e.g., attention and cognitive control networks) and connectivity across the brain at large. We propose to examine “connectomes” integrated over all of resting brain activity (i.e. “static” or averaged connectivity), in accord with the state of the art connectivity studies published in the field over the past decade, including most recent methodological advances. This proposal is a complement to, and is complemented by other proposals under consideration that focus on dynamic connectivity and multimodal classification analysis. With respect to the dynamic connectivity proposal, (i.e. changes in connectivity “states” and metrics of state dwell-time and transitions over the resting epoch), it is yet unknown what approach will yield most informative findings, whether these will be additive or overlapping, and even methodologically it will be highly valuable to compare PTSD findings from both approaches. In relationship to the proposal to use multi-modal MRI measures (including metrics of cortical thickness, subcortical volumetrics, diffusor tension imaging (DTI), and resting state connectivity) as features in machine-learning classification, our proposed analysis is focused only on, but is also far more comprehensive re: rs fMRI analytics. It will produce complementary metrics of integrated network connectivity that can be used in subsequent machine-learning classification studies proposed. This parallel processing analytic approach will thereby greatly increase efficiency and collaboration between ENIGMA-PTSD member groups. Below we briefly describe scientific rational and the specific methods of our proposal.
Large-Scale ICNs in Trauma-related Psychopathology Resting state functional connectivity (rsFC) analyses have identified multiple large-scale intrinsic connectivity networks (ICNs), that correspond to task-related connectivity patterns8,9. Over the past 5-7 years, several research groups, including ours, have reported ICN seed-based studies showing individuals with PTSD have hyper-connectivity within the nodes of Salience Network (SN) linked to detection of threat, salience and integration of interoceptive, and emotional information, like insula, dorsal anterior cingulate (dACC) and in some models, amygdala10-13, Several seed-based studies have also reported decreased within-network connectivity in the “task-negative” Default Mode Network (DMN, linked to self-referential processing, autobiographical memory, and “mind wandering”: posterior cingulate cortex (PCC), ventral medial PFC (vmPFC) / subgenual PFC (sgPFC), and hippocampus12-17 .Finally, PTSD has been associated with aberrantly increased DMN-SN (e.g. DMN-amygdala and insula) cross-network connectivity i.e. network desegregation12,15,18. We have previously hypothesized increased DMN-SN connectivity at rest as potentially underlying PTSD hyperarousal symptoms, associated with intrusive, automatic distress reactions to both external cues and internal physiological states.
A few seed-based studies also reported that other ICNs may also contribute to PTSD pathophysiology. Hypervigilance and concentration problems in PTSD involve inappropriate attention to threat and distractibility, which could involve aberrant cross-network hyperconnectivity between Dorsal Attention Network (DAN, involved in voluntary deployment of attention and externally-directed cognitions) and Ventral Attention Network (VAN, involved in reorientation to unexpected events) with DMN and/or SN19,20. Recent studies also suggest PTSD related dysfunction in Frontal-Pariental Control Network (FPCN, executive functions and “top-down” control emotional regulation) including impaired modulation of DMN by FPCN21-23 and FPCN-SN hyperconnectivity18, consistent with long-standing evidence of dysfunction in frontal cortex and “top-down” inhibitory processing and emotional regulation in PTSD.
Thus while a reasonably circumscribed set of findings had been emerging from seed based connectivity studies, not all of the hypotheses were replicated by all the studies, and large N definitive study is needed to identify these findings that are reliably replicable in the majority of PTSD patients.
In addition to seed-based analyses, connectome-wide studies using graph theory analysis suggest PTSD-related alterations in the global architecture of brain connectivity. For example, decreased “small-world” topology, consistent with a failure of equilibrium between functional segregation and integration processes in PTSD26,27 However, PTSD small-world findings to date are mixed28,29 as are reports of PTSD alterations in clustering coefficient28,30. We also present preliminary connectomic data utilizing joint Independent Components Analysis (King, Block, Sripada, & Liberzon, unpublished data) that finds PTSD associated with altered expression of large-scale components of connectivity characterized by increased desegregation (aberrantly increased connectivity) between DMN, DAN, and VAN (see preliminary data below). Clearly, well powered studies are required to test these global connectivity findings.
Hypotheses: Based upon previous work from the Michigan group10,12,20,23 and others11,13,15,16,18,21,22, we propose two overarching hypotheses to be tested:
H1. PTSD is associated with Increased within-network connectivity in the Salience Network; decreased within-network connectivity in the Default Mode Network; and increased between-network connectivity (desegregation) between the Salience Network and Default Mode Network
H2: PTSD is associated with altered connectivity across the whole brain and also in attention and cognitive control networks, including increased connectivity between DAN and DMN, DAN and SN, and decreased connectivity between FPCN and DMN
Specific Aims: To test our primary hypotheses, we propose the following Specific Aims:
Aim 1: Confirmatory Seed-based Analyses will be used to test a priori PTSD connectivity hypotheses regarding SN and DMN in seed-based analyses utilizing canonical seed ROIs within critical nodes of relevant ICNs (SN: dorsal ACC, insula, DMN: PCC, vmPFC, hippocampus).
1a. Seed-to-voxel analyses testing whole-brain functional connectivity to canonical seed ROIs.
1b. ROI-to-ROI network analyses testing connectivity between a priori DMN and SN nodes.
Aim 2. Exploratory Whole-Connectome Analyses We will utilize three complementary approaches based on analyses of “connectomes” (i.e., all pairwise connections between all “nodes” encompassing the entire brain) to identify network abnormalities associated with PTSD in an unbiased, whole-brain level of analysis at three levels of analysis: testing at the level of (a) individual “edges” between connectome nodes, (b) large scale components of connectivity, and (c) global network characteristics.
2a. Connectome-based Modeling will be used to identify specific sets of individual edges within connectomes that are statistically different between PTSD and healthy controls using whole-brain statistical thresholding, and categorize them by canonical networks.
2b. Joint Independent Components Analyses (jICA) will be used to identify PTSD-related differences in large-scale components of connectivity networks. jICA will reveal individual-level distributed network components of connectivity (both within and between canonical ICN networks) that show cohesive patterns of variability between individuals, and allow us to identify effects of PTSD on network components in group-level comparisons.
2c. Graph theory analyses will be used to identify PTSD-related alterations in global topological properties of neural networks, including functional segregation and integration (e.g., metrics of global efficiency, small-worldness, modularity, etc.)
We have preliminary seed-based and connectome-based analyses from three medium-sized studies with 3T fMRI resting state data (total of N=147 OEF/OIF combat veterans with and without PTSD and N=40 healthy community controls): (1) a locally funded PTSD fMRI study (PI Israel Liberzon, MD, co-PI Anthony King, PhD) with N=15 OEF/OIF combat PTSD, and N=14 OEF/OIF combat healthy controls, (2) PTSD Mindfulness-based Exposure Therapy study (W81XWH-08-2-0208, PI: Israel Liberzon, MD; co-PI Anthony King, PhD) with N=39 OEF/OIF combat PTSD patients and N=26 healthy controls, and the PROgRESS study (W81XWH-11-1-0073, coordinating PI: Sheila Rauch, PhD, fMRI PI: Israel Liberzon, MD) with N=69 OEF/OIF combat PTSD and N=29 OEF/OIF combat healthy controls.
Connectomic Analysis reveals PTSD associated with altered expression of a connectivity component comprised of negative connectivity (segregation) between DMN, DAN, and VAN.
Joint ICA of N=32 OEF/OIF PTSD and N=26 healthy controls: PTSD Associated with Desegregation of DMN and DAN and VAN. “Connectomes” were generated from Pearson correlations of spatially-averaged time-series of all pair-wise combination of 1068 regions of interest (ROIs) in a 12 mm grid spanning the whole brain. Connectome data were decomposed into 15 spatial components using joint Independent Components Analysis (jICA) and ordered into categories of ICNs (e.g. DMN, SN, DAN, FPCN/ CEN, etc.). Expression of a joint ICA “parietal” independent component (IC) was significantly decreased in PTSD compared to healthy controls (t(48.4) = 3.6, p = 0.0008). This parietal IC was expressed medially and bilaterally and was composed of negative cross-network connectivity of the DAN and VAN to DMN; consistent with PTSD showing relative desegregation between DMN, DAN, and VAN. Expression of this IC was associated with a behavioral measure, mean response time (RT) in the Attention Network Task (ANT) t(51) = -2.1378, p = 0.04 (PTSD had lower expression of the parietal IC and higher RT (slower responses) in the ANT).
Participants: We will utilize resting state fMRI scans and clinical and demographic metadata collected by the ENIGMA-PGC PTSD consortium from ~3000 participants; approximately N=1500 with current PTSD diagnosis, and N=1500 healthy people who have been exposed to trauma but no current or lifetime history of PTSD. PTSD diagnosis and severity and demographics will be obtained from individual studies; the majority of studies use CAPS to assess PTSD, covariates will be harmonized across studies as necessary. Participants will be excluded if they have lifetime diagnosis of schizophrenia, bipolar disorder, or other psychotic disorder, significant cognitive disability, or neurological disorders. Participants with organic brain damage due to traumatic brain injury will be excluded, but mild TBI / post-concussive syndrome will be allowed but coded as a co-variable. Presence of major depression will be allowed in PTSD patients but not healthy controls but will be coded as a covariable in analyses. Other covariables of interest will include childhood maltreatment / adversity and adult trauma exposure categories, harmonized across studies.
MRI Data Acquisition, pre-processing, and artifact removal. Resting state raw fMRI data contributed from all sites will be pre-processed using the standardized multisite ENIGMA pipeline to process resting state scans in a harmonized manner. We will use the single-modality ENIGMA rsfMRI preprocessing pipeline (recently reported in Adhikari et al., 2018)24 utilizing model-free Marchenko-Pastur PCA based de-noising, and which will apply standard preprocessing steps (e.g. spatial distortion removal, realignment, co-registration, warping to MNI space). Standard artifact removal will involve linear detrending and regression applied to each voxel’s time series to remove nuisance effects. Regressors will include the top five PCs from PCA of BOLD time series from cerebrospinal fluid and white matter masks (to remove signal from cardiac and respiratory cycles, and the time‐series band‐ pass filtered in the 0.01 – 0.10 Hz range, and motion terms (from realignment), their first derivatives and quadratic terms (to reduce motion artifact). Each methodological approach also includes methods for removing motion artifact, which is particularly important for rs fMRI analyses. To reduce motion artifact that can confound connectivity measures, we will apply “Motion Scrubbing” (censoring of individual frames from the time series). We will initially set the framewise displacement threshold for excessive motion to 0.2 mm; subjects with more than 33% of frames removed by scrubbing will be excluded.
Aim 1. Confirmatory Seed-based Approaches. To test a priori hypotheses regarding DMN and SN connectivity, we will test “seed-to-voxel” and ROI-to-ROI network connectivity analyses of the CONN toolbox utilizing previously validated a priori “seed” ROIs from canonical nodes in DMN (PCC, vmPFC, bilateral hippocampus) and SN (bilateral insula, dorsal ACC, bilateral amygdala), and whole-brain statistical thresholding (FDR). The models will be controlled for nuisance variables (e.g. site, length of scan, etc.) and the potential influence of covariates of interest including sex, age, childhood trauma / adversity, adult trauma load, mild TBI, depression co-morbidity, etc. will also be tested and controlled for in models of PTSD.
Aim 2 Exploratory Connectome-based Approaches to test whole-brain connectome-wide hypotheses:
Connectome generation. We will have produce a whole-brain resting state functional connectome as wehave previously reported32,33, placing ROIs encompassing nineteen 3 x 3 x 3 mm voxels in a regular grid spaced at 12 mm intervals throughout the brain, yielding 1080 ROIs in total. We will utilize the network parcellation of Yeo et al., 20119, which included convergent methods to test reliability and parcellation using grid-based connectomic methods similar to current study. Spatially averaged time series will be extracted from each of ROIs followed by regression to remove nuisance effects. Pearson’s correlation coefficients will be calculated pairwise between time courses for each ROI, producing a cross‐correlation map with ~500,000 non‐redundant entries, and Fisher’s r-z‐transformation is applied for subsequent statistical analysis, as we have previously reported32,33,34,35
2a. Connectome-based Tests. All edges in connectivity matrices of generated connectomes will be regressed to the group variable (PTSD diagnosis or severity), and a whole-brain significance threshold applied to select the PTSD-related edges36. Contrasts will be applied to detect group differences in functional connectivity, two-tailed t tests on the values of each cell to create a “difference matrix” of t scores representing the magnitude of group difference at each edge. We then will threshold the matrix and apply multiple comparison corrections using network-based statistics (NBS)37, in which the largest fully connected network of suprathreshold edges, or “component,” is identified, and its extent is defined as the number of edges it comprises. Group assignments of subjects will be randomly permuted (10,000 permutations) to create a null distribution for the expected component size due to chance36. The data generated from these analyses can also be used in predictive modeling of PTSD; i.e., selected edges can be summarized to a per subject single value, and a predictive model built that classifies a set of edges to a diagnostic group. Summary values calculated for each subject in the testing set will be input to predictive model, and cross-validation (e.g. leave-one-out, k-fold cross validation) can be applied.
2b. Joint Independent Components Analysis (jICA). We will identify patterns of inter‐individual variation across the connectome as we have reported previously using jICA32,33,34,35 to identify independent components (ICs) that exhibit cohesive patterns of variation across subjects. “ICA-Fix” will reduce movement noise regressing 24 movement parameters (6 parameter time series, temporal derivatives, and squares). We will use GIFT toolbox, mialab.mrn.org/ software/gift/) for jICA and empirically choose models (e.g. 10-20 ICs) to best represent brain areas of networks of interest (DMN, FPCN, etc) without over-parcellation. Time courses will be a matrix of N volumes x k ICs, and will be post-processed using GIFT toolbox: lowpass ﬁltering (0.15 Hz), detrending, and “despiking” (AFNI 3dDespike algorithm replaces data points larger than median deviation with a spline ﬁt; similar to “data scrubbing” but does not disrupt temporal continuity. Pearson correlations will be calculated between ICs using the processed data, resulting in functional connections that will be fit in an IC correlation matrix ordered into networks (DMN, SN etc.), and individual variability will be back-projected to test PTSD-related differences in strength of derived connectivity networks in between-subjects analyses.
2c. Graph Theory metrics. We will utilize graph theory algorithms to calculate both global (small-worldness, global efficiency, strength, modularity) and local network metrics (local efficiency, node degree and betweenness centrality) from individual-level connectomes38,39. Because networks of individual subjects differ in the number of edges, we will employ sparsity thresholds (S) to the correlation matrices to provide each graph (network) with the same number of edges39, defined as the fraction of the total number of edges remaining in a network. We will utilize threshold strategies to produces networks that could estimate small-worldness with sparse properties and the minimum possible number of spurious edges. We will cаlculate areа under the curve (AUC) for each network metric (over the sparsity range of thresholds) as a summarized scalar for the topological characterization of neural networks38,39, and will utilize nonparametric permutation tests to identify significant between-group differences in the AUCs of all of the graph metrics (e.g. network efficiency, small-world properties and nodal characteristics) of the generated functional connectomes between the PTSD patients and healthy control subjects.
1. Etkin A, Wager TD. Functional neuroimaging of anxiety: a meta-analysis of emotional processing in PTSD, social anxiety disorder, and specific phobia. The American journal of psychiatry. 2007;164(10):1476-1488.
2. O’Doherty DC, Chitty KM, Saddiqui S, Bennett MR, Lagopoulos J. A systematic review and meta-analysis of magnetic resonance imaging measurement of structural volumes in posttraumatic stress disorder. Psychiatry research. 2015;232(1):1-33.
3. Stark EA, Parsons CE, Van Hartevelt TJ, et al. Post-traumatic stress influences the brain even in the absence of symptoms: A systematic, quantitative meta-analysis of neuroimaging studies. Neuroscience and biobehavioral reviews. 2015;56:207-221.
4. Alexopoulos GS, Hoptman MJ, Kanellopoulos D, Murphy CF, Lim KO, Gunning FM. Functional connectivity in the cognitive control network and the default mode network in late-life depression. Journal of affective disorders. 2012;139(1):56-65.
5. Anticevic A, Cole MW, Murray JD, Corlett PR, Wang XJ, Krystal JH. The role of default network deactivation in cognition and disease. Trends Cogn Sci. 2012;16(12):584-592.
6. Cole MW, Repovs G, Anticevic A. The frontoparietal control system: a central role in mental health. Neuroscientist. 2014;20(6):652-664.
7. Hamilton JP, Chen MC, Gotlib IH. Neural systems approaches to understanding major depressive disorder: an intrinsic functional organization perspective. Neurobiology of disease. 2013;52:4-11.
8. Fox MD, Snyder AZ, Vincent JL, Corbetta M, Van Essen DC, Raichle ME. The human brain is intrinsically organized into dynamic, anticorrelated functional networks. Proc Natl Acad Sci U S A. 2005;102(27):9673-9678.
9. Yeo BT, Krienen FM, Sepulcre J, et al. The organization of the human cerebral cortex estimated by intrinsic functional connectivity. J Neurophysiol. 2011;106(3):1125-1165.
10. Sripada RK, King AP, Garfinkel SN, et al. Altered resting-state amygdala functional connectivity in men with posttraumatic stress disorder. Journal of psychiatry & neuroscience : JPN. 2012;37(4):241-249.
11. Rabinak CA, Angstadt M, Welsh RC, et al. Altered amygdala resting-state functional connectivity in post-traumatic stress disorder. Frontiers in psychiatry. 2011;2:62.
12. Sripada RK, King AP, Welsh RC, et al. Neural dysregulation in posttraumatic stress disorder: evidence for disrupted equilibrium between salience and default mode brain networks. Psychosomatic medicine. 2012;74(9):904-911.
13. Herringa RJ, Birn RM, Ruttle PL, et al. Childhood maltreatment is associated with altered fear circuitry and increased internalizing symptoms by late adolescence. Proc Natl Acad Sci U S A. 2013;110(47):19119-19124.
14. Bluhm R, Williamson P, Lanius R, et al. Resting state default-mode network connectivity in early depression using a seed region-of-interest analysis: decreased connectivity with caudate nucleus. Psychiatry and clinical neurosciences. 2009;63(6):754-761.
15. Lanius RA, Bluhm RL, Coupland NJ, et al. Default mode network connectivity as a predictor of post-traumatic stress disorder symptom severity in acutely traumatized subjects. Acta psychiatrica Scandinavica. 2010;121(1):33-40.
16. Miller DR, Hayes SM, Hayes JP, Spielberg JM, Lafleche G, Verfaellie M. Default Mode Network Subsystems are Differentially Disrupted in Posttraumatic Stress Disorder. Biol Psychiatry Cogn Neurosci Neuroimaging. 2017;2(4):363-371.
17. Birn RM, Patriat R, Phillips ML, Germain A, Herringa RJ. Childhood maltreatment and combat posttraumatic stress differentially predict fear-related fronto-subcortical connectivity. Depress Anxiety. 2014;31(10):880-892.
18. Rabellino D, Tursich M, Frewen PA, et al. Intrinsic Connectivity Networks in post-traumatic stress disorder during sub- and supraliminal processing of threat-related stimuli. Acta psychiatrica Scandinavica. 2015;132(5):365-378.
19. Poole VN, Robinson ME, Singleton O, et al. Intrinsic functional connectivity predicts individual differences in distractibility. Neuropsychologia. 2016;86:176-182.
20. Russman Block S, King AP, Sripada RK, Weissman DH, Welsh R, Liberzon I. Behavioral and neural correlates of disrupted orienting attention in posttraumatic stress disorder. Cogn Affect Behav Neurosci. 2017;17(2):422-436.
21. Clausen AN, Francisco AJ, Thelen J, et al. PTSD and cognitive symptoms relate to inhibition-related prefrontal activation and functional connectivity. Depress Anxiety. 2017;34(5):427-436.
22. Tursich M, Ros T, Frewen PA, Kluetsch RC, Calhoun VD, Lanius RA. Distinct intrinsic network connectivity patterns of post-traumatic stress disorder symptom clusters. Acta psychiatrica Scandinavica. 2015;132(1):29-38.
23. King AP, Block SR, Sripada RK, et al. Altered Default Mode Network (DMN) Resting State Functional Connectivity Following a Mindfulness-Based Exposure Therapy for Posttraumatic Stress Disorder (PTSD) in Combat Veterans of Afghanistan and Iraq. Depress Anxiety. 2016;33(4):289-299.
24. Adhikari BM, Jahanshad N, Shukla D, et al. Heritability estimates on resting state fMRI data using ENIGMA analysis pipeline. Pac Symp Biocomput. 2018;23:307-318.
25. Veraart J, Novikov DS, Christiaens D, Ades-Aron B, Sijbers J, Fieremans E. Denoising of diffusion MRI using random matrix theory. Neuroimage. 2016;142:394-406.
26. Jung, W. H., Chang, K. J., & Kim, N. H. (2016). Disrupted topological organization in the whole-brain functional network of trauma-exposed firefighters: A preliminary study. Psychiatry Research: Neuroimaging, 250, 15-23.
27. Du, M. Y., Liao, W., Lui, S., Huang, X. Q., Li, F., Kuang, W. H., … & Gong, Q. Y. (2015). Altered functional connectivity in the brain default-mode network of earthquake survivors persists after 2 years despite recovery from anxiety symptoms. Social cognitive and affective neuroscience, 10(11), 1497-1505.
28. Zhang, X. D., Yin, Y., Hu, X. L., Duan, L., Qi, R., Xu, Q., … & Li, L. J. (2017). Altered default mode network configuration in posttraumatic stress disorder after earthquake: A resting-stage functional magnetic resonance imaging study. Medicine, 96(37), e7826.
29. Suo, X., Lei, D., Chen, F., Wu, M., Li, L., Sun, L., … & Gong, Q. (2016). Anatomic insights into disrupted small-world networks in pediatric posttraumatic stress disorder. Radiology, 282(3), 826-834
30. Lei, D., Li, K., Li, L., Chen, F., Huang, X., Lui, S., … & Gong, Q. (2015). Disrupted functional brain connectome in patients with posttraumatic stress disorder. Radiology, 276(3), 818-827
31. Suo, X., Lei, D., Li, K., Chen, F., Li, F., Li, L., … & Gong, Q. (2015). Disrupted brain network topology in pediatric posttraumatic stress disorder: a resting‐state fMRI study. Human brain mapping, 36(9), 3677-3686.
32. Kessler D, Angstadt M, Sripada C. Growth Charting of Brain Connectivity Networks and the Identification of Attention Impairment in Youth. JAMA Psychiatry. 2016;73(5):481-9.
33. Sripada CS, Kessler D, Angstadt M. Lag in maturation of the brain’s intrinsic functional architecture in attention-deficit/hyperactivity disorder. Proc Natl Acad Sci U S A. 2014;111(39):14259-64.
34. Sripada C, Kessler D, Fang Y, Welsh RC, Prem Kumar K, Angstadt M. Disrupted network architecture of the resting brain in attention-deficit/hyperactivity disorder. Hum Brain Mapp. 2014;35(9):4693-705.
35. Sripada CS, Kessler D, Welsh R, Angstadt M, Liberzon I, Phan KL, Scott C. Distributed effects of methylphenidate on the network structure of the resting brain: a connectomic pattern classification analysis. Neuroimage. 2013; 81:213-21.
36. Shen X, Finn ES, Scheinost D, Rosenberg MD, Chun MM, Papademetris X, Constable RT. Using connectome-based predictive modeling to predict individual behavior from brain connectivity. Nat Protoc. 2017;12(3):506-518.
37. Zalesky A, Fornito A, Bullmore ET. Network-based statistic: identifying differences in brain networks. NeuroImage, 2010; 53:1197-207.
38. Hosseini SM, Hoeft F, Kesler SR. GAT: a graph-theoretical analysis toolbox for analyzing between-group differences in large-scale structural and functional brain networks. PLoS One. 2012;7(7):e40709
39. Liang X, Wang J, Yan C, Shu N, Xu K, Gong G, et al. (2012) Effects of Different Correlation Metrics and Preprocessing Factors on Small-World Brain Functional Networks: A Resting-State Functional MRI Study. PLoS ONE 7(3): e32766.