Auditory verbal hallucinations (AVHs) refer to one's sensory experience of hearing voices even in the absence of any external auditory stimuli, which is a core positive symptom of schizophrenia and usually affects 70-75% of schizophrenic patients (Nayani and David, 1996). As seen in previous studies (Ohtani et al., 2014; Wagner et al., 2015), AVH patients show structural abnormalities and functional dysfunction mainly concentrated in brain regions including the bilateral temporal lobe, prefrontal lobe and parietal cortex (Hugdahl et. al.,2017). AVHs showed activation in auditory cortex in the absence of external sounds, but when AVHs were given external sounds, the activation of these brain regions did not increase, but decreased (Kompus et al.,2011). So, the AVH is regarded as a result of the functional breakdown in the monitoring of inner speech generation or attention bias towards internal speech at the expense of orienting to external space (Allen et al., 2007) (Kompus et al.,2011). However, the pathophysiology of AVHs is still poorly understood, and there is no consensus in those reported mechanisms for AVHs.

In healthy controls, P300 is evoked by target stimuli during the visual or auditory oddball task (Squires et al., 1975) and is usually characterized by the largest positive peak at approximately 300 ms after target onset and prominently distributed in the parietal region. In essence, the target stimuli can evoke a clear P300 only if the related information is efficiently processed in the brain, which is attributed to the effective allocation of related resources. Previous studies, however, showed that patients with schizophrenia exhibited abnormalities in cognitive function (e.g., sensory processing, working memory, attention, thinking and decision making), as well as less efficient allocation of resources, compared to healthy controls (Kahn and Keefe, 2013; Leitman et al., 2010; Li et al., 2018; Smucny et al., 2013). Consequently, reduced P300 amplitude is regarded as an endophenotype of schizophrenia (Bramon et al., 2005; Bramon et al., 2004) and can thus be used to index the neurobiological vulnerability of schizophrenia, as well as to evaluate the cognitive capacity of schizophrenic patients (Leitman et al., 2010; Rissling et al., 2010). In previous studies, P300 has been used to discriminate schizophrenic patients from healthy controls (Li et al., 2019b; Turetsky et al., 2015). For example, using multiple P300 variables (e.g., four amplitudes and three latencies), Chun and colleagues (2013) obtained 83% classification accuracy when recognizing schizophrenic patients (Chun et al., 2013). However, using P300 amplitude or latencies to discriminate different schizophrenic symptoms is still far from satisfactory, and discriminating AVH patients from non-AVH patients is also very difficult. This may because both AVH and non-AVH groups experience cognitive impairments. During an oddball task, the P300 mainly evaluates the participants’ response to target stimuli, leaving the inter-regional information exchanges unveiled (Li et al., 2019c), this may be a way to differentiate different types of schizophrenic patients.

The human brain usually works as a large-scale complex network consisting of spatially distributed but functionally linked brain regions (Bassett and Sporns, 2017; Li et al., 2016; Li et al., 2019a). Efficient information transfer and processing have been proven to rely on the structural and functional brain network (Li et al., 2015; Zhang et al., 2015; Zhang et al., 2016). As postulated by the "dysconnectivity hypothesis of schizophrenia", especially the functional connectivity between prefrontal and temporal regions, some schizophrenic phenomena are best understood in terms of abnormal interactions between different areas of the brain at the levels of physiology, functional anatomy and cognition (Friston and Frith, 1995; Stephan et al., 2009). Cognitive capacity deficits are also seen in both AVH and non-AVH patients. In our previous studies (Li et al., 2019b), brain network analysis was successfully applied in classifying schizophrenic patients from healthy controls, and the highest accuracy of 90.48% was achieved by using the spatial network topologies of resting and task brain states. However, these studies mainly focused on the comparison between schizophrenic patients and healthy controls, and no research has been conducted to investigate possible differences in the brain network between AVH and non-AVH patients, which will provide evidence for better clinical diagnosis, including early screening for AVHs and better pharmacological treatment for schizophrenic patients with AVH.

In our current study with a specially designed auditory oddball task, brain network analysis was conducted to investigate the cognitive capacity deficits of both AVH and non-AVH schizophrenic patients, as well as their potential network differences. It was hypothesized that the potential network differences between the two patient groups would help the discrimination of AVH patients from non-AVH patients. We would thus be able to classify AVH and non-AVH patients appropriately. Furthermore, it should also be possible to quantitatively evaluate and compare the corresponding performance on the auditory oddball task of these two types of schizophrenic patients in the canonical electrophysiological features, i.e., P300 amplitudes and power spectral density (PSD), providing possibilities for early diagnosis of patients with AVH and better pharmacological treatment.