Please use this identifier to cite or link to this item: http://localhost:8081/xmlui/handle/123456789/15166
Title: FUNCTIONAL NETWORK CONNECTIVITY ANALYSIS OF HUMAN BRAIN
Authors: Sariya, Yogesh Kumar
Keywords: Alzheimer;Epilepsy;Autism Spectrum;Autism Spectrum Disorder (ASD)
Issue Date: Jul-2018
Publisher: I.I.T Roorkee
Abstract: Nowadays, the neurological disorders such as Alzheimer, epilepsy, autism spectrum disorder, Parkinson’s disease, multiple sclerosis etc. has increased to an alarming level. Autism Spectrum Disorder (ASD), in this series, is an umbrella term for multiple neurodevelopmental conditions characterized by repetitive or stereotyped behaviors and pervasive deficits in social communications and interactions. The ASD is considered a lifelong disability which has an impact on both the individual and the family, as well as being a cost to society in general. Among these costs are additional health care, disability support in school and, in some instances, the loss of a productive working life and the provision of social security. By 2017, estimates of the prevalence of autism by the world health organization were 1 per 160 children, more than a 30-fold increase from the first studies of autism prevalence. From the perspective of social healthcare, it’s an utmost requirement to understand their etiology. Functional integration of the brain networks on a macroscopic level is being analyzed exhaustively to establish the biomarkers of neurological disorders. It is also termed as functional network connectivity analysis. During the past decade, the disrupted connectivity theory has generated considerable interest as a pathophysiological model for ASD. This theory postulates that deficiencies in the way the brain coordinates and synchronizes activity among different regions may account for the clinical symptoms of ASD. The most common version of this hypothesis proposes that individuals with ASD have weak connections between distant brain regions and increased connections within local regions. The existing connectivity data in ASD are inconsistent and one possible explanation for the variability is that although altered connectivity is a pathogenic mechanism, there are insufficient specificity in existing hypotheses, insufficient precision in the techniques, excessive sensitivity to confounds, or insufficient power in studies to correctly identify the supporting evidence. The functional network connectivity in ASD significantly fluctuated in the research works carried out in this field because of methodological and subject choice contrasts. Early examinations regularly centered around locale differences in activation during tasks, with more recent studies utilizing resting state functional MRI concentrated in seed-based techniques and low-order ICA models. Developmental changes in functional connectivity have received inadequate attention and the discrepancies between findings of autism related hypoi connectivity and hyper-connectivity might be reconciled by taking developmental changes into account as per the age range of the persons. The present work emphasizes to bridge the inconsistencies in the literature and tries to establish a reliable biomarker which truly governs the signature manifestations of ASD patients. A resting-state examination is preferred rather a task-based one, because the imaging protocol is typically faster and the collected data serves multiple mapping purposes, thus fitting better into the usually limited patient scanning schedule. To accomplish this, the objectives of the present research work are formulated as: (1) comparing the ICA algorithms on the basis of their abilities to decompose the fMRI images, (2) building standard 3D templates for the naming of intrinsic connectivity networks, (3) age-stratified assessment of social and cognitive dysfunction of ASD through functional network connectivity (FNC), and (4) agestratified functional network-based dynamic connectivity analysis in autism with higher order ICA model. A variety of existing ICA algorithms have been implemented so far for fMRI images. With a view that algorithms that are overlooked may outperform the most opted, a comparative study is taken up in the first objective to analyze their abilities for the purpose of decomposition of fMRI images. In this work, ten independent component algorithms: Fast ICA, INFOMAX, SIMBEC, JADE, ERICA, EVD, RADICAL, ICA-EBM, ERBM, and COMBI are compared. Their separation abilities are adjudged on both, synthetic and real fMRI images. Performance to decompose synthetic fMRI images is being monitored on the basis of spatial correlation coefficients, time elapsed to extract independent components and the visual appearance of independent components. Ranking of their performances on task-based real fMRI images are based on the closeness of time courses of identified independent components with model time course and the closeness of spatial maps of components with spatial templates while their competencies for resting state fMRI data are analyzed by examining how distinctly they decompose the data into the most consistent resting state networks. Sum of mutual information between all the permutations of decomposed components of resting state fMRI data is also calculated. Based on all the acquired results, it is deduced that ERICA, EVD, and SIMBEC are not suitable algorithms for decomposition of fMRI images. Aggregate observations reveal that it is not worthwhile to use ERBM or ICA-EBM sepaii rately but their combination is a better choice to use for fMRI decomposition. RADICAL is exceptionally sluggish and thus it is not a good choice. Among the overlooked algorithms and most opted algorithms, COMBI is a better choice for fMRI decomposition. The COMBI is fastest as well as it decomposes components quite distinctively. The ICA finds the autonomous components of fMRI signal that represent a mix of ’true’ brain networks and artifactual components from various sources such as cerebrospinal fluid, white matter, blood vessels and head motion. The separation of decomposed components into these two groups currently remains a semi-manual process, determined by quantitative metrics but reliant on the experience of the neuro-physicians. The labeling of RSN is being done either by utilizing the spatial correlation between the given 2D layout and component images or in light of the premise of ROIs effectively reported as the captivating anatomical part of those in past literary works. Splitting of components due to model order ambiguity, moderate advance in computer vision, higher likelihood of confusion of segments those are spread over more than one projection of the brain and repeating ROIs for different RSNs are the significant obstacles in utilizing 2D RSN layouts for the marking reason. To overcome this inadequacy, 3D templates are proposed in the second objective with the end goal of making RSN recognition automated. Proposed 3D templates are a superior substitute being free from these shortcomings. The use of volumetric overlap of decomposed RSN with 3D templates instead of the spatial correlation between RSN and 2D templates is expected to give more accurate results. These templates are developed by overlaying the manually labeled RSNs on 3D glass brain. These are selected from 100 decomposed components. The anatomic positions of them are based on the results of Talairach client where the toolbox was commanded to search in a cube of +/- 2mm. Images of manually labeled RSN are co-registered to sample image of multi-image analysis GUI (MANGO) toolbox (http://ric.uthscsa.edu/mango/) followed by the surface rendering. They are rendered on 3D glass brain for the purpose of better visibility. These 3D templates can be used as the standard for the labeling of RSN and are perfectly suitable for RsfMRI studies employing lower/higher order ICA model. Any RSN will get the name of the parent template which encompasses it and the quantification can be done based on the similarity indexes that measures the volumetric overlap viz., Dice coefficient or Jaccard coefficient. iii Static functional network connectivity (sFNC) assessment to explore the limited cognitive ability of autistic subjects is executed in the third objective. The sFNC among six cognitive brain networks, viz. anterior default mode network, posterior default mode network, two frontoparietal networks (LFPN and RFPN), basal ganglia network (BG) and salience network (SN) have been examined. To understand the developmental trajectory of ASD, functional magnetic resonance imaging (fMRI) dataset of autistic children, adolescents and adults are considered. Constrained maximal lag correlation between each pair of networks of interest by calculating Pearsons correlation and constraining the lag between the time courses. The number of possible pair-wise combinations to examine between-network connectivity is 15 with six networks of interest. Subject-specific time-courses were detrended and despiked, then filtered using a fifth-order Butterworth low-pass filter with a cutoff frequency of 0.15 Hz. The SN manifest aberrant patterns of brain connectivity in the various stages of developmental trajectory. Attention allocation to stimuli, salient to the individual, is a conventional responsibility of the SN and atypical development of the SN may lessen interest in social interaction which is a signature characteristic of ASD. For the underlying two formative stages, the two frontoparietal networks are not connected and thus no availability of connectivity between them for the atypical population. This may be one of the reasons behind their hallmark behavioral characteristics. The fourth objective of this study is to explore a whole brain dynamic functional network connectivity differences in a developmental trajectory. The dynamic FNC are evaluated utilizing a sliding window approach instantiated in the dFNC toolbox in GIFT toolbox. First, the time-courses were detrended and despiked using 3D despike in the AFNI software followed by filtering using a fifth-order Butterworth low-pass filter with a cutoff frequency of 0.15 Hz. Then, FNC covariance matrices were calculated between all pairwise RSNs for each subject using the correlations derived from previously done ICA analysis by moving a Gaussian window in 1 TR (time of repetition) increments across the subject time-courses. Successive FNC matrices for each window were then concatenated to form an array [number of RSNs x number of RSNs x (number of window units)] representing a state transition vector, or how the FNC state changed through time for each subject. Subsequently, a clustering analysis is done to examine the structure and frequency of FNC patterns that recurred in the state transiiv tion vectors. The k-means clustering algorithm was applied to the individual arrays of FNC covariance matrices using the City method and the algorithm iterated a maximum of 200 times before convergence. The results of this objective are dissimilar to the speculations that hyper-connectivity of brain networks are prevalent in young children with ASD, while hypoconnectivity are more common in young people and adults with the disorder when compared to typically developing cohorts. The statistically significant functional network connectivity differences (fdr<= 0.05) are sparse in the group of children and adults and even the significant intra-connectivity differences do not influence over the interconnectivity differences in the children’s group of ASD.
URI: http://localhost:8081/xmlui/handle/123456789/15166
Research Supervisor/ Guide: Anand, R.S.
metadata.dc.type: Thesis
Appears in Collections:DOCTORAL THESES (Electrical Engg)

Files in This Item:
File Description SizeFormat 
G28718.pdf26.34 MBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.