Introducing a novel approach of network oriented analysis of ERPs, demonstrated on adult attention deficit hyperactivity disorder

Abstract

Objective

Introducing a network-oriented analysis method (brain network activation [BNA]) of event related potential (ERP) activities and evaluating its value in the identification and severity-grading of adult ADHD patients.

Methods

Spatio-temporal interrelations and synchronicity of multi-sited ERP activity peaks were extracted in a group of 13 ADHD patients and 13 control subjects for the No-go stimulus in a Go/No-go task. Participants were scored by cross-validation against the most discriminative ensuing group patterns and scores were correlated to neuropsychological evaluation scores.

Results

A distinct frontal–central–parietal pattern in the delta frequency range, dominant at the P3 latency, was unraveled in controls, while central activity in the theta and alpha frequency ranges predominated in the ADHD pattern, involving early ERP components (P1–N1–P2–N2). Cross-validation based on this analysis yielded 92% specificity and 84% sensitivity and individual scores correlated well with behavioral assessments.

Conclusions

These results suggest that the ADHD group was more characterized by the process of exerting attention in the early monitoring stages of the No-go signal while the controls were more characterized by the process of inhibiting the response to that signal.

Significance

The BNA method may provide both diagnostic and drug development tools for use in diverse neurological disorders.

Introduction

Anatomical, physiological and clinical evidence has indicated that functional networks spanning spatial and temporal scales form a critical feature of information processing in the brain (Bullmore and Sporns, 2009). More specifically, the coordination of widely distributed and parallel brain activity, which is correlated with sensory processing, has been suggested to rise as a function of the synchronization of network potential oscillations (Engel et al., 2001, Buzsaki and Draguhn, 2004). This synchronization has been suggested to govern the effective strength of connections between regions in the brain, providing a flexible mechanism for the generation of diverse functional networks, within the framework of the more rigid anatomical network (Fries, 2005). Importantly, these mutual influences between neuronal groups, based on rhythmic activity, were found to be state-specific in the frequency, time and location domains (Womelsdorf et al., 2007).

The existing methods of complex network analysis, which attempt to characterize functional connection patterns, invoke different types of interdependencies such as temporal correlations (Stam, 2005, Stam et al., 2007, Zhou et al., 2009, Rubinov and Sporns, 2010), coherence across brain regions (Sun et al., 2004, Salvador et al., 2005; Murias et al., 2007; Zhou et al., 2009) and synchronization likelihood between electrodes (Ahmadlou and Adeli, 2010). Other approaches to characterize large-scale networks (Stam et al., 2007, Bullmore and Sporns, 2009, Rubinov and Sporns, 2011) use different aspects of graph theoretical analyses (De Vico Fallani et al., 2008) including small world networks (Bassett and Bullmore, 2006), multiscale data mining (Eldawlatly et al., 2009), and interdependencies between and within community structures (Ahmadlou and Adeli, 2011). In addition, signal decomposition (e.g., PCA, ICA) (e.g., Mueller et al., 2010, Mueller et al., 2011) and windowing or segmentation (e.g., “microstate analysis”, Koenig and Lehmann, 1996) have also been used as methods for pattern extraction.

A common denominator across most of these methods is that individual measures based on bivariate or multivariate synchronizations are used to characterize patterns of functional integration and segregation within and between brain regions or community structures. In contrast, our approach, named brain network activation (BNA), a new methodology for EEG analysis, compares between entire networks, each representative of a different subject group or condition, without restricting analysis to sub-parts of the brain neuronal network.

The BNA algorithm, in an unsupervised manner, seeks and unravels task-related structures in the time–location–frequency space of the EEG, by projecting individual data points derived from ERP waveforms into clusters. For each experimental state, one or more unique patterns common to all (or most) subjects in an experimental group or state are revealed, to which individual activity patterns may then be compared (see Section 2 and Appendix A). The comparison is made to all elements of the ensuing group pattern, rather than to a discrete extracted parameter (i.e., complexity, connectivity, coherence, etc.), as done in other methods; while microstate analysis results in topographical maps that distinguish between groups or conditions (e.g., Brunet et al., 2011) the BNA method compares between group-common multidimensional patterns. The basic building block of each group pattern is the coupling of discrete events (each in a specific location and frequency band) into pairs, based on the inter-subject synchronicity of the time-lapse between the events. In addition, the BNA methodology enables the comparison between the network obtained at the group level and the network of an individual subject, and to score the degree of congruity between the two.

Several recent studies suggest that the pathophysiology of ADHD is related to abnormal brain activity at the functional network level (reviewed in Konrad and Eickhoff (2010)). These findings reflect the change in perspective that has occurred in etiological models of ADHD that shifted the focus of the pathology from regional brain deficiencies to dysfunctions at distributed networks. However, most of the studies undertaking the network approach involved children or adolescents and many of them were resting-state functional magnetic resonance imaging (fMRI) or diffused tensor imaging studies (Makris et al., 2007, Konrad and Eickhoff, 2010). ERP studies of adults with ADHD are scarce (Bekker et al., 2005, Fallgatter et al., 2005, Prox et al., 2007, McLoughlin et al., 2010) and very few ADHD studies addressed widespread functional networks in the brain using ERPs, either in children (e.g., Ahmadlou and Adeli, 2011) or in adults (e.g., Mueller et al., 2010, Mueller et al., 2011).

The purpose of this work was to evaluate whether the BNA methodology may provide some new insights regarding ADHD at the functional network level in adult patients. Converging neuropsychological results point at impaired inhibitory control as a fundamental deficit of ADHD in children (Pliszka et al., 2000, Yordanova et al., 2001, Johnstone et al., 2007, Wodka et al., 2007) as well as in adults (Fallgatter et al., 2005, Prox et al., 2007). Therefore, we chose the Go/No-go task (Bokura et al., 2001, Liddle et al., 2001, Yordanova et al., 2001, Garavan et al., 2002) to compare between the networks of adult ADHD patients and healthy controls. Using the Go/No-go paradigm during EEG recording, the cortical manifestation of neural mechanisms involved in attentive and inhibitory processes can be studied with a high degree of temporal resolution (Manuel et al., 2010, Smith and Douglas, 2011).

The current diagnosis of ADHD in the clinic relies on a set of behavioral measures and a subjective clinical assessment, while a neurophysiologically-based measure for the diagnosis and rating of ADHD in the individual subject may contribute to clinical practice. Towards this end, subjects were assigned individual scores based on the degree of conformity of their own brain activity to the ensuing discriminative group activity patterns. In addition, specificity and sensitivity of the use of these scores for classification as well as their correlation with behavioral measures were determined.