NeuroMark: a fully automated ICA method to identify effective fMRI markers of brain disorders

Abstract

Increasing sharing initiatives on neuroimaging data have provided unprecedented opportunities to study brain disorders. Standardized approaches for capturing reproducible and comparable biomarkers are greatly needed. Here, we propose a method, NeuroMark, which leverages a priori-driven independent component analysis to effectively extract functional brain network biomarkers from functional magnetic resonance imaging (fMRI) data. NeuroMark automatically estimates features adaptable to each individual and comparable across subjects by taking advantage of the replicated brain network templates extracted from 1828 healthy controls as guidance to initialize the individual-level networks. Four studies including 2454 subjects were conducted spanning six brain disorders (schizophrenia, autism spectrum disorder, depression, bipolar disorder, mild cognitive impairment and Alzheimer's disease) to evaluate the proposed method from different perspectives (replication, cross-study comparison, subtle difference identification, and multi-disorder classification). Results demonstrate the great potential of NeuroMark in its feasibility to link different datasets/studies/disorders and enhance sensitivity in identifying biomarkers for patients with challenging mental illnesses. Significance StatementIncreasing evidence highlights that features extracted from resting fMRI data can be leveraged as potential biomarkers of brain disorders. However, it has been difficult to replicate results using different datasets, translate findings across studies, and differentiate brain disorders sharing similar clinical symptoms. It is important to systematically characterize the degree to which unique and similar impaired patterns are reflective of brain disorders. We propose a fully automated method (called NeuroMark) that leverages priori-driven independent component analysis (ICA) using replicated brain network templates to estimate individual-subject network features. Evaluated by four studies involving six different brain disorders, we show that NeuroMark can effectively link the comparison of biomarkers across different studies/datasets/disorders and enable classification between complex brain disorders, while also providing information about relevant aspects of whole brain functional connectivity.

Aberrant Dynamic Functional Connectivity of Default Mode Network in Schizophrenia and Links to Symptom Severity

“…A covariance matrix was calculated to measure the dFC (Figure 1 Step2). The dFC estimates of each window for each subject were concatenated to form a (C × C × T) array (where C = 7 denotes the number of subnodes, and T = 124 in COBRE and T = 137 in FBIRN denotes the number of windows), which represented the changes in brain connectivity between subnodes as a function of time (Allen et al, 2014;Calhoun et al, 2014;Fu et al, 2019). Since the temporal resolution and the eye condition of the two datasets were different, we did not combine them in our study and chose to analyze them separately instead.…”

A deep learning fusion model for brain disorder classification

“…In this work, we employed multiple brain spatial functional networks, functional connectivity matrix, and brain structural morphology volume as the inputs. Specifically, for each subject, we estimated 53 brain functional networks and a functional network connectivity (FNC) matrix from resting-state fMRI data using our previously developed fully automated NeuroMark approach [28]. The data-driven independent component analysis (ICA) method is able to extract brain functional features with both individual-subject specificity and inter-subject correspondence, by taking advantaging of the network templates from a large sample of population as priors and a multiple objective optimization method [29].…”

Deep Temporal Organization of fMRI Phase Synchrony Modes Promotes Large-Scale Disconnection in Schizophrenia

“…Hence, we adopted the 50 aggregate networks (estimated from a 100-component 8 group spatial ICA analysis on 405 subjects in Allen et al (2014), and implemented constrained spatial ICA (Lin et al, 2010) on the preprocessed data of every participant, using the Group ICA of fMRI Toolbox (GIFT 9 ). Notably, the artifactual components (i.e., the physiological, head movement, and imaging artifact components) had already been identified and excluded in Allen et al (2014), such that the remaining 50 functional parcels comprise sub-components of reproducible large-scale restingstate networks (Kiviniemi et al, 2009;Abou-Elseoud et al, 2010;Allen et al, 2011;Du et al, 2019). These networks include the subcortical (SC), auditory (AUD), somatomotor (SM), visual (VIS), cognitive control (CC), default mode network (DMN), and cerebellum (CB) ( Figure 2C).…”