An approach to directly link ICA and seed-based functional connectivity: Application to schizophrenia

Highlights

• An enhanced ICA is proposed to parcellate an ultra-large functional connectivity matrix in whole brain voxel pairs.

• The proposed cmICA links two methods for functional connectivity analysis, i.e. ICA-based analysis and seed-based analysis.

• The cmICA is used to investigate functional connectivity differences between 60 SZ patients and 61 healthy controls.

• The detected dysconnectivities are found to be highly associated with cognitive deficits in schizophrenia patients.

Abstract

Independent component analysis (ICA) and seed-based analyses are widely used techniques for studying intrinsic neuronal activity in task-based or resting scans. In this work, we show there is a direct link between the two, and show that there are some important differences between the two approaches in terms of what information they capture. We developed an enhanced connectivity-matrix independent component analysis (cmICA) for calculating whole brain voxel maps of functional connectivity, which reduces the computational complexity of voxel-based connectivity analysis on performing many temporal correlations. We also show there is a mathematical equivalency between parcellations on voxel-to-voxel functional connectivity and simplified cmICA. Next, we used this cost-efficient data-driven method to examine the resting state fMRI connectivity in schizophrenia patients (SZ) and healthy controls (HC) on a whole brain scale and further quantified the relationship between brain functional connectivity and cognitive performances measured by the Measurement and Treatment Research to Improve Cognition in Schizophrenia (MATRICS) battery. Current results suggest that SZ exhibit a wide-range abnormality, primarily a decrease, in functional connectivity both between networks and within different network hubs. Specific functional connectivity decreases were associated with MATRICS performance deficits. In addition, we found that resting state functional connectivity decreases was extensively associated with aging regardless of groups. In contrast, there was no relationship between positive and negative symptoms in the patients and functional connectivity. In sum, we have developed a novel mathematical relationship between ICA and seed-based connectivity that reduces computational complexity, which has broad applicability, and showed a specific application of this approach to characterize connectivity changes associated with cognitive scores in SZ.

Introduction

The diagnosis of schizophrenia has been traditionally made based on symptoms, e.g., hallucinations, delusions etc. However the schizophrenic phenotype varies drastically in terms of patterns and severities across individual cases (Jablensky, 2010), and could be significantly suppressed by antipsychotic treatments and other factors (Lui et al., 2010). Researchers have been searching for neurophysiologically specific biomarkers/causes. Among numerous techniques and findings, resting state functional connectivity from neuroimaging has emerged as a promising candidate (Arbabshirani et al., 2017; Friston, 2011; Sheffield and Barch, 2016).

Functional connectivity during the resting state provides a novel and complementary tool to explore brain organization (Biswal et al., 2010; Fox et al., 2005; Fox et al., 2007) as well as the mechanics of brain function and behavior (Baird et al., 2005; Van Dijk et al., 2010). It reveals more reliable “intrinsic” diagrams of brain neuronal communications, compared to task-based analysis given that the latter induces less than 5% energy changes (Raichle and Gusnard, 2002). This intrinsic functional connectivity has shown increasing evidence in capturing the variability associated with neuropsychiatric illnesses, e.g., schizophrenia (Camchong et al., 2011; Fitzsimmons et al., 2013; Garrity et al., 2007; Lynall et al., 2010; Skudlarski et al., 2010). Resting state data collection also reduced performance confounds in SZ with cognitive deficits (Friston, 2011). Because of this, resting state functional connectivity has been widely studied to predict disease states (Arbabshirani et al., 2013; Arbabshirani et al., 2017; Craddock et al., 2009; Du et al., 2012; Krishnadas et al., 2014; Shen et al., 2010; Silva et al., 2014; Venkataraman et al., 2012).

The disconnection/dysconnectivity hypothesis of schizophrenia has been considered for over 20 years (Friston, 1998; Friston and Frith, 1995; Pettersson-Yeo et al., 2011; Stephan et al., 2006; Stephan et al., 2009) where the term “dysconnectivity” refers to either abnormal reductions in functional interactions or to abnormal increases (Stephan et al., 2009). Although functional dysconnectivity has been found in many studies, the revealed patterns show a great deal of inconsistency across studies (Sheffield and Barch, 2016). Alterations between cortical and subcortical regions have been reported (Cetin et al., 2014; Cheng et al., 2015; Damaraju et al., 2014; Sui et al., 2015; Woodward et al., 2012; Zhou et al., 2010). Other frequent findings are hyper- or hypo-connectivity associated with default mode networks (DMN), with both intra-DMN and inter-DMN-nonDMN connectivity (Karbasforoushan and Woodward, 2012; Moran et al., 2013; Unschuld et al., 2014). Changes in global (whole brain) functional connectivity have also been found in some studies (Bassett et al., 2012; Cole et al., 2011; Lynall et al., 2010). One suggestion to address this heterogeneity in findings is that cognitive dysfunction may contribute to functional connectivity (Sheffield and Barch, 2016). Detecting links between cognitive impairments, a core feature of schizophrenia, and aberrant functional connectivity may better help understand the neurobiological mechanism underlying disconnection and perhaps help clarify the diagnostic boundaries of the syndrome.

Methodologically speaking, it is helpful to consider two types of functional connectivity methods primarily used at present – seed-based temporal correlation analysis and spatial independent component analysis (ICA) (Beckmann et al., 2005; Buckner and Vincent, 2007; Calhoun and Adali, 2012; Calhoun et al., 2001a; Calhoun et al., 2009b; Cole et al., 2010; Joel et al., 2011; Van Den Heuvel and Pol, 2010). Seed-based approaches find the connectivity of a seed to the rest of the brain. The seed can be a collection of points based on prior functional magnetic resonance imaging (fMRI) studies or be based on an atlas, in which case it is a seed with a larger region-of-interest (ROI). The advantage of this method lies in its simplicity, which pinpoints directly the voxel regions to which the seed region is connected. However, seed-based methods are hypothesis driven which requires prior knowledge of seeds or ROIs whose representativeness is not always reliable (Cole et al., 2010; Sohn et al., 2015). The information of a functional connectivity map from this type of method is restricted and biased to a selected seed region, leaving undetermined the connectivity patterns on a whole-brain scale (Van Den Heuvel and Pol, 2010). ICA on the other hand is data driven and computationally efficient in examining general connectivity patterns from whole brain. ICA yields brain voxel regions with strong temporal coherence, seen as intra-connected maps. However, the spatial network/connectivity maps identified by ICA are derived by maximizing statistical independence from each other, and its interpretation in terms of connectivity coherence may not be as straight-forward as seed-based results (Sheffield and Barch, 2016). The ICA approach may in some cases only detect localized synchronous voxels in a partial intrinsic network (e.g., the split regions of default mode network) especially when using high model orders (Calhoun and de Lacy, 2017). To study connectivity across different networks, ICA is usually employed in a component/regional macroscopic scale often with a focus on inter-network connectivity, called functional network connectivity or FNC (Allen et al., 2012a; Jafri et al., 2008). It would be very useful to be able to utilize the computationally efficient and other benefits of ICA in the context of a whole-brain seed-based approach.

One goal of this paper is to develop an efficient data-driven method to examine resting state functional connectivity in SZ on a whole brain scale, and quantify the relationship between brain functional dysconnectivity with both cognitive deficits and symptomatology. Voxel-based connectivity maps have been studied to look for brain connectivity differences in schizophrenia with diffusion imaging and with resting-state fMRI data (Cheng et al., 2015; Skudlarski et al., 2010; Wu et al., 2015). These voxel-based connectivity maps have high dimension $\left(NxN\right)$, where N is the number of brain voxels. This is not only computationally expensive, but a direct comparison of these connectivity maps across groups can lead to underestimation of important differences after correcting for multiple comparisons. The cmICA method (Wu et al., 2015) was developed as one way to address this problem, and it has been previously applied to parcellate whole brain diffusion imaging data (Wu et al., 2015; Wu et al., 2013). In that paper, we show that cmICA can be used for dual parcellation of the brain connectivity (regardless of it being structural or functional). The dual parcellation consists of a set of spatially independent maps S and a corresponding dual set of spatial maps R, such that R defines the brain connectivity magnitudes from independent sources S to the whole brain (see Discussion). Here we propose a computationally optimized cmICA for functional connectivity data. We apply the enhanced cmICA to resting state fMRI data collected from 60 SZ and 61 HC. The new method increases sensitivity to subtle connectivity changes across subjects (Allen et al., 2011; Koch et al., 2010; Wu et al., 2015). Moreover, our approach increases longer-ranged inter-regional connectivity sensitivity and spatial localization compared to conventional ICA connectivity methods (Wu et al., 2015), both of which may facilitate the detection of relationships between abnormal connectivity and cognitive deficits in schizophrenia.