A Statistical Approach in Human Brain Connectome of ... - IEEE Xplore

0 downloads 0 Views 751KB Size Report
and controls family wise error rates (in weak sense). The major regions with significantly reduced interconnecting fiber volume or average tract length were ...
A Statistical Approach in Human Brain Connectome of Parkinson Disease in Elderly People Using Network Based Statistics Mohammad Hadi Aarabi*, Aida Kamalian1, Bahram Mohajer, Mahdi Shirin Shandiz, Ehsan Eqlimi, Ahmad Shojaei and Hamidreza Safabakhsh 

Abstract— Parkinson's Disease (PD) is a progressive neurodegenerative disorder assumed to involve different areas of CNS and PNS. Thus, Diffusion Tensor Imaging (DTI) is used to examine the areas engaged in PD neurodegeneration. In the present study, we computed average tract length and fiber volume as a measure of white matter integrity and adopted Network Based Statistics (NBS) to conduct group analyses between age- and gender-matched PD patients and healthy control connectivity matrices. NBS is a powerful statistical tool that utilizes the presence of every link in connectivity matrices and controls family wise error rates (in weak sense). The major regions with significantly reduced interconnecting fiber volume or average tract length were cingulum, temporal lobe, frontal lobe, parahippocampus, hippocampus, olfactory lobe, and occipital lobe.

Keywords- diffusion tensor imaging (DTI), parkinson disease (PD), structure brain connectivity and network based statistics (NBS). I. INTRODUCTION The structure of brain is extensively explained as an integrated, complex network using concepts of graph theory, in recent years. The mentioned approach is an attempt to study and quantify whole brain as a single graph, conceiving it as main functioning sections (nodes or vertices) linked by interconnecting fibers (edges)[1]. One of the known models to define brain network's organization is Small-world network, a model paradoxically possessing high levels of local connectivity (cliqueness) and global integration at once[2]. High level of local connectivity is the key feature of 'hubs' which are the structural centers of information processing. Communicating fibers enable rapid exchange of information between hubs, hence conferring on brain the ability of working as a fast, multitasking unit_ the so-called global integration. Variations of global integration and local connectivity levels underlying various neurological disorders have been areas of recent vigorous studies [3]. A recent study has suggested selective damage of hubs in neurological disorders. In case of Alzheimer's disease, temporal lobe hub-concentrated lesions were noted [4]. *M. H. Aarabi is with the Students' Scientific Research Center,Tehran University of Medical Sciences,Tehran,Iran and Basir Eye Health Research Center, Tehran, Iran. (corresponding author : [email protected]) A. Kamalian, B. Mohajer, Ehsan Eqlimi and M. S. Shandiz are with the Students' Scientific Research Center,Tehran University of Medical Sciences,Tehran,Iran. A.Shojaei and H.Safabakhsh are with the Basir Eye Health Research Center, Tehran, Iran. 1 Co-authors who contributed equally.

978-1-4244-9270-1/15/$31.00 ©2015 IEEE

Parkinson's disease (PD) is a progressive neurodegenerative disorder mainly characterized by motor symptoms including resting tremor, rigidity, bradykinesia, and postural instability. The poor quality of life caused by motor symptoms deteriorate by emergence of non-motor symptoms especially dementia. Non-motor manifestations of PD comprise olfactory malfunction, amnesia, impaired visuospatial ability, depression, and sleep disturbances [5]. Case-control studies using fMRI has illuminated some areas of damage[6], however, although functional imaging gives us a clue about functionally impaired areas, structural examination of PD-associated damage leads us to a more specific understanding of pathologic brain connectivity especially in white matter areas. Diffusion Tensor model utilizes diffusion-specific signal intensity reduction of water in Diffusion Weighted images to quantify diffusion constant of water in at least six directions. Estimating the dominant direction of water diffusion is the next step leading to DTI tractography of white matter. Previously conducted region of interest studies (ROI) assess and compare diffusion parameters such as MD and FA of healthy and patient groups in certain regions [7]. In other approaches, the whole brain is simplified as graphs. Connectivity matrices are subsequently calculated for various parameters (e.g. MD, FA, or ƛ). However, these matrices were compared indirectly through assessing features of the graphs e.g. clustering coefficient and average shortest path length in order to investigate brain connectivity network in pathologic conditions [3]. In our study, we aimed to calculate healthy and PD whole brain connectivity matrices for average tract length and average fiber volume and compare them directly through Network Based Statistics (NBS) [8]. By adopting NBS no more topological property interpretation is needed, familywise error rate is reduced, and every structure exhibited by the connections is utilized to yield a greater accuracy in detection of pathologically valuable connections. Our study is determined to discover main areas of parkinsonian brain damage not only from a statistical viewpoint but also through exploring two diffusion parameters less addressed in previous researches. Average fiber volume and average tract length are used as demonstrations of atrophy and damage levels in brains of parkinsonian patients. II. PROCEDURE A. Participants Participants involved in this research were recruited from Parkinson's Progression Markers Initiative (PPMI). DWI images were obtained for 18 patients (7 F 11M, mean age

4310

73.33) and 12 controls (4F 8M, mean age 71.50). Participants were tested and confirmed negative for any neurological disorders apart from PD. The participants' PD status was confirmed by Movement Disorder Society-Unified Parkinson's Disease Rating Scale (MDS-UPDRS) and the loss of dopaminergic neurons were observed in DaTscans. Every participant involved in this research has signed informed written consents in order to share their unidentified clinical data to investigators. B. Data Acquisition Data used in the preparation of this paper was obtained from Parkinson's Progression Markers Initiative (PPMI) database (www.ppmi-info.org/data/) [9]. This dataset was acquired on a 3 Tesla Siemens scanner, producing 64 DWI (repetition time = 7748 ms, echo time = 86 ms; voxel size: 2.0×2.0×2.0 mm3; field of view = 224×224 mm) at b = 1000 s/mm2 and one b0 image along with a 3D T1-weighted structural scan (repetition time = 8.2 ms, echo time = 3.7 ms; flip angle = 8˚, voxel size: 1.0×1.0×1.0 mm3; field of view = 240mm, acquisition matrix = 240 ×240).

presence of every link is utilized to yield greater power than a generic procedure to control FWER. III. RESULTS T-test statistics were applied with threshold of 2.9 and significant results (p-values for average fiber volume and tract length were 0.0202 and 0.02892, respectively) were obtained (Tables I and II). The hubs exhibiting significantly reduced average tract length of interconnecting fibers mainly include: cingulum, parahippocampus, temporal lobe, parietal lobe, palladum, putamen, and occipital lobe. A significant decrease of interconnecting fiber volume was observed in the same mentioned regions, as well as olfactory lobe and insula. Tables I and II illustrate the detailed tracts. In order to display the tracts comprehensively, a graphic illustration is shown in Fig. 1 and 2.

C. Diffusion MRI Data Processing, Network Construction and Group Analysis The DWI data were analyzed and processed in ExploreDTI [10]. The diffusion tensor was estimated using a weighted linear least square regression procedure proposed in [11]. For subject motion correction and eddy-current induced geometrical distortion, DWI data were rigidly registered with MNI atlas[12]. During this processing step, we adjusted the B-matrices with appropriate reorientations and included the required signal intensity modulation with the Jacobian determinant of the spatial transformation[13]. Whole-brain tractography was performed using a deterministic algorithm. Fibers were reconstructed by placing seed points on a uniform grid across the data set at 2 mm isotropic resolution and by following the main diffusion direction (as defined by the principal eigenvector) unless the fiber tract entered a voxel with a FA 30°). The step size was set at 0.5 mm. The whole-brain fiber tracts were then parcellated using the automated anatomical labeling (AAL), a widely used atlas to derive nodes in graph theoretical analyses of neuroimaging data [14]. The mentioned atlas divides brain into 116 cortical and subcortical regions (58 for each hemisphere including cerebellum). The inter-regional connectivity between the 116 ROIs demarcated on the AAL template was computed by applying the ROI masks to the reconstructed fiber tracts using ExploreDTI. In this fashion we determined the volume and average tract length that originated in one ROI (i) and terminated in another one (j), for all 116 ROIs defined on the atlas, creating a 116 x 116 connectivity matrix (CM). In the next step we used NBS for group analysis [8]. NBS is a method based on the principles underlying traditional cluster-based thresholding of statistical parametric maps to control family-wise error rate (in the weak sense) when massunivariate testing is performed at every connection of the graph. The gain in power offered by NBS increases as the number of nodes and links is increased, which serves well in our complex neural network. In brief, by using NBS, the 4311

TABLE I.

VOLUME OF FIBERS

Pathway Frontal_Sup_Orb_L to Frontal_Mid_R Frontal_Mid_R to Frontal_Inf_Tri_R Frontal_Inf_Oper_R to Frontal_Inf_Tri_R Frontal_Inf_Oper_R to Insula_R Frontal_Inf_Tri_R to Insula_R Cingulum_Post_L to Cingulum_Post_R Cingulum_Post_L to Hippocampus_L Cingulum_Ant_R to Hippocampus_R Olfactory_R to ParaHippocampal_L Frontal_Inf_Orb_R to Occipital_Sup_L Cuneus_L to Occipital_Mid_R Occipital_Sup_L to Occipital_Mid_R Occipital_Inf_R to Parietal_Sup_R Frontal_Inf_Tri_L to Precuneus_L Cingulum_Post_L to Precuneus_L Cingulum_Post_R to Precuneus_L Cingulum_Post_R to Precuneus_R Hippocampus_R to Precuneus_R Occipital_Mid_R to Precuneus_R Frontal_Inf_Orb_R to Putamen_R Parietal_Sup_R to Pallidum_L Precuneus_L to Pallidum_L Heschl_L to Temporal_Sup_L Olfactory_R to Temporal_Pole_Sup_L Cingulum_Post_R to Temporal_Pole_Sup_L Precuneus_L to Temporal_Pole_Sup_L Pallidum_R to Temporal_Pole_Sup_L Heschl_L to Temporal_Pole_Sup_L Insula_R to Temporal_Pole_Sup_R Putamen_R to Temporal_Pole_Sup_R

Test Stat 2.96 3.07 3.86 3.68 2.92 3.13 3.10 3.48 3.13 2.99 3.53 3.04 3.33 2.91 3.06 3.00 2.99 3.12 3.55 2.94 3.28 3.14 3.15 3.09 2.95 3.00 3.27 3.01 3.13 3.18

TABLE II.

AVERAGE TRACT LENGTH

Pathway Supp_Motor_Area_L to Cingulum_Post_L Frontal_Sup_L to Hippocampus_L Cingulum_Post_L to ParaHippocampal_L Frontal_Inf_Orb_R to Occipital_Sup_L Occipital_Mid_L to Occipital_Mid_R ParaHippocampal_L to Fusiform_L Frontal_Inf_Orb_L to Parietal_Sup_L Frontal_Inf_Orb_R to Parietal_Sup_R Cingulum_Post_R to Parietal_Sup_R Occipital_Mid_L to Parietal_Sup_R Occipital_Inf_R to Parietal_Sup_R Hippocampus_L to Precuneus_L ParaHippocampal_L to Precuneus_L Fusiform_L to Precuneus_L Parietal_Sup_R to Precuneus_L Parietal_Sup_R to Putamen_L Parietal_Sup_R to Pallidum_L Frontal_Inf_Orb_L to Pallidum_R Rolandic_Oper_L to Heschl_L Postcentral_L to Heschl_L Cingulum_Ant_L to Temporal_Pole_Sup_L Hippocampus_L to Temporal_Pole_Sup_L ParaHippocampal_L to Temporal_Pole_Sup_L Precuneus_L to Temporal_Pole_Sup_L Pallidum_R to Temporal_Pole_Sup_L Heschl_L to Temporal_Pole_Sup_L ParaHippocampal_L to Temporal_Inf_L Parietal_Sup_R to Vermis_3

Test Stat 3.20 2.98 2.96 4.06 2.98 3.01 4.02 3.57 3.63 3.10 3.10 3.17 2.92 2.97 3.06 4.41 3.81 3.02 3.64 3.62 3.64 3.20 3.39

Figure 1. Graphical illustration of significantly altered connectome pathway: volume of fibers

3.26 3.43 3.37 2.92 3.05

IV. CONCLUSION In this study, for the first time, we explored PD connectome exploiting NBS in group analysis instead of studying brain network in the framework of topological properties of connectivity matrices. 116x116 weighted connectivity matrices were constructed for average tract length and average fiber volume to locate damaged tracts underlying motor and non-motor manifestations of PD. Treating brain as a network and seeking the underlying reason of its disturbances in the connections of network is a perspective gaining more and more popularization. Modeling brain as connectivity matrices has the advantage of investigating brain totally and as a unit without becoming time-consuming. Network Based Statistics is a powerful statistical tool which takes every single structural connection into account. Accordingly, the risk of eliminating valuable connections, which is an inevitable outcome of summarizing

a whole network in its topological properties, is remarkably reduced. Whereas quite powerful, NBS has limitations due to the fact that it can only provide weak control of the family-wise error rate. Simply put as a reduced ability to reject the null hypothesis. Plus, with NBS, the value of the threshold used to define the set of suprathreshold links is an arbitrary choice. NBS is very helpful only if it is utilized cautiously in suitable networks (e.g. NBS is ineffective when the links associated with the contrast or effect of interest are isolated and do not form any connected structures). Its main disadvantage though is the massive number of multiple comparisons that must be performed. Another potential pitfall threatening the validity of our results is the crucial yet largely arbitrary matter of node selection[15]. Undoubtedly the results may vary when the location and number of nodes change. The mentioned problem was controlled by utilizing widely used, anatomically reliable AAL parcellation. Our analyses confirm the significance of most regions proposed in previous researches. Putamen and palladum are structures comprising basal ganglia, among first structures discovered to be damaged in PD brain[16]. Tracts interconnecting cingulum, parahippocampal gyrus, temporal lobe, frontal lobe, precuneus, insula, and hippocampus are supported through functional MRI, DTI, and voxel-based morphometery analysis studies[17, 18]. Tracts projecting from olfactory lobe and occipital lobe were barely discussed in preceding papers, despite several studies suggesting olfactory malfunction and occipital-lobe-related sleep disturbances as manifestations of damage to the mentioned areas[19, 20].

4312

[4]

[5] [6]

[7]

[8] [9] [10]

[11]

Figure 2. Graphical illustration of significantly altered connectome pathway: average tract length

In summary, the current findings suggest interconnecting fibers between cingulum, temporal lobe, parahippocampus, olfactory lobe, frontal lobe, etc. as main sites of atrophy and damage when brain is accurately examined without disregarding any structural connection using a powerful statistical tool. If these findings are extended to a larger scale, this readily available approach may have utility in developing noninvasive differential diagnosis or region targeted therapies.

[12]

[13] [14]

[15]

ACKNOWLEDGMENT PPMI – a public-private partnership – is funded by the Michael J. Fox Foundation for Parkinson’s Research and funding partners, including [list the full names of all of the PPMI funding partners found at www.ppmiinfo.org/fundingpartners]. This research is also supported by Tehran University of Medical Sciences 94-01-61-28328. The authors thank Dr.Pasalar for her kindly support of this work. REFERENCES [1] [2]

[3]

[16]

[17]

[18]

[19]

O. Sporns, D. R. Chialvo, M. Kaiser, and C. C. Hilgetag, "Organization, development and function of complex brain networks," Trends Cogn Sci, vol. 8, pp. 418-25, Sep 2004. S. Achard, R. Salvador, B. Whitcher, J. Suckling, and E. Bullmore, "A Resilient, Low-Frequency, Small-World Human Brain Functional Network with Highly Connected Association Cortical Hubs," The Journal of Neuroscience, vol. 26, pp. 63-72, January 4, 2006 2006. K. T. Olde Dubbelink, A. Hillebrand, D. Stoffers, J. B. Deijen, J. W. Twisk, C. J. Stam, et al., "Disrupted brain network topology

[20]

4313

in Parkinson's disease: a longitudinal magnetoencephalography study," Brain, vol. 137, pp. 197-207, Jan 2014. W. de Haan, K. Mott, E. C. van Straaten, P. Scheltens, and C. J. Stam, "Activity dependent degeneration explains hub vulnerability in Alzheimer's disease," PLoS Comput Biol, vol. 8, p. e1002582, 2012. P. J. Garcia-Ruiz, K. R. Chaudhuri, and P. Martinez-Martin, "Non-motor symptoms of Parkinson's disease A review...from the past," J Neurol Sci, vol. 338, pp. 30-3, Mar 15 2014. H. Yang, X. J. Zhou, M. M. Zhang, X. N. Zheng, Y. L. Zhao, and J. Wang, "Changes in spontaneous brain activity in early Parkinson's disease," Neurosci Lett, vol. 549, pp. 24-8, Aug 9 2013. Z. Zheng, S. Shemmassian, C. Wijekoon, W. Kim, S. Y. Bookheimer, and N. Pouratian, "DTI correlates of distinct cognitive impairments in Parkinson's disease," Hum Brain Mapp, vol. 35, pp. 1325-33, Apr 2014. A. Zalesky, A. Fornito, and E. T. Bullmore, "Network-based statistic: identifying differences in brain networks," Neuroimage, vol. 53, pp. 1197-207, Dec 2010. K. Marek, D. Jennings, S. Lasch, A. Siderowf, C. Tanner, T. Simuni, et al., "The parkinson progression marker initiative (PPMI)," Progress in neurobiology, vol. 95, pp. 629-635, 2011. A. Leemans, B. Jeurissen, J. Sijbers, and D. Jones, "ExploreDTI: a graphical toolbox for processing, analyzing, and visualizing diffusion MR data," in 17th Annual Meeting of Intl Soc Mag Reson Med, 2009, p. 3537. Q. Collier, J. Veraart, B. Jeurissen, A. J. den Dekker, and J. Sijbers, "Iterative reweighted linear least squares for accurate, fast, and robust estimation of diffusion magnetic resonance parameters," Magnetic Resonance in Medicine, pp. n/a-n/a, 2014. G. K. Rohde, A. S. Barnett, P. J. Basser, S. Marenco, and C. Pierpaoli, "Comprehensive approach for correction of motion and distortion in diffusion-weighted MRI," Magn Reson Med, vol. 51, pp. 103-14, Jan 2004. A. Leemans and D. K. Jones, "The B-matrix must be rotated when correcting for subject motion in DTI data," Magnetic Resonance in Medicine, vol. 61, pp. 1336-1349, 2009. N. Tzourio-Mazoyer, B. Landeau, D. Papathanassiou, F. Crivello, O. Etard, N. Delcroix, et al., "Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain," Neuroimage, vol. 15, pp. 273-89, Jan 2002. A. Zalesky, A. Fornito, I. H. Harding, L. Cocchi, M. Yucel, C. Pantelis, et al., "Whole-brain anatomical networks: does the choice of nodes matter?," Neuroimage, vol. 50, pp. 970-83, Apr 15 2010. M. Skorpil, V. Soderlund, A. Sundin, and P. Svenningsson, "MRI diffusion in Parkinson's disease: using the technique's inherent directional information to study the olfactory bulb and substantia nigra," J Parkinsons Dis, vol. 2, pp. 171-80, 2012. K. Ito, Y. Masutani, K. Kamagata, H. Yasmin, Y. Suzuki, K. Ino, et al., "Automatic extraction of the cingulum bundle in diffusion tensor tract-specific analysis: feasibility study in Parkinson's disease with and without dementia," Magn Reson Med Sci, vol. 12, pp. 201-13, 2013. D. Zhang, X. Liu, J. Chen, and B. Liu, "Distinguishing patients with Parkinson's disease subtypes from normal controls based on functional network regional efficiencies," PLoS One, vol. 9, p. e115131, 2014. J. G. Goldman, G. T. Stebbins, V. Dinh, B. Bernard, D. Merkitch, L. deToledo-Morrell, et al., "Visuoperceptive region atrophy independent of cognitive status in patients with Parkinson's disease with hallucinations," Brain, vol. 137, pp. 849-59, Mar 2014. E. Y. Lee, P. J. Eslinger, G. Du, L. Kong, M. M. Lewis, and X. Huang, "Olfactory-related cortical atrophy is associated with olfactory dysfunction in Parkinson's disease," Mov Disord, vol. 29, pp. 1205-8, Aug 2014.