Brain Structural and Functional Neuroimaging Features are Associated With Improved Auditory Hallucinations in Patients With Schizophrenia After Real-Time fMRI Neurofeedback

Abstract

Auditory hallucinations (AHs) are debilitating and often treatment-resistant symptoms of schizophrenia (SZ). Real-time functional magnetic resonance imaging (fMRI) neurofeedback (NFB) is emerging as a flexible brain circuit-based tool for targeting AH via self-modulation of brain activity. A better understanding of what baseline characteristics predict NFB success will enhance its clinical utility. Previous work suggests that AH symptomology implicates measures across multiple modalities, including T1 structural MRI (sMRI), diffusion-weighted MRI (dMRI), and resting-state fMRI (rsfMRI). Specifically, AH severity and treatment response are associated with thinner superior temporal gyrus (STG), thinner dorsolateral prefrontal cortex (DLPFC), reduced white matter integrity in tracts connecting brain regions implicated in SZ symptomatology, increased within-default mode network (DMN) connectivity, and reduced DMN-DLPFC anticorrelation. In this study, we tested the individual and combined contributions of multimodal brain features for the prediction of AH change after NFB in adults (N = 25, 36.1 ± 10.0 years, 24% females) with SZ spectrum disorders (SZ or schizoaffective disorder) and frequent medication-resistant AH. Participants underwent a baseline MRI scan (including sMRI, dMRI, and rsfMRI) and were randomly assigned to receive NFB from their STG (n = 12, real condition) or NFB from their motor cortex (MC) (n = 13, sham condition). NFB success was operationalized as the improvement in AH severity after NFB. We found that higher baseline AH severity, greater STG thickness, decreased dorsal cingulum integrity, increased within-DMN resting-state functional connectivity, and increased DMN-DLPFC anticorrelation were each individually correlated with reduction in AH severity. However, in a combined regression model, DMN-DLPFC connectivity emerged as the only independent variable that explained the unique variance in AH change. These results suggest that a specific rsfMRI measure, namely DMN-DLPFC connectivity, may be a promising predictor of NFB success in reducing AH and support the precision medicine approach. Trial Registration: ClinicalTrials.gov identifier: NCT03504579.

Figure 1

Mindfulness-based neurofeedback experiment setup. MRI scanner: The task display was projected and reflected in a mirror attached to the scanner head coil. Real-time fMRI data analysis: During the task, fMRI data were transferred and analyzed to determine the STG activation level when voice recordings were played. On neurofeedback runs, the activation level was displayed within the task interface. Task interface: Participants were shown a cue to either attend to their own voice or ignore a stranger's voice. Participants were then played the corresponding voice recordings. Participants were asked to provide a self-rating on their listening performance. On neurofeedback runs, activation level was displayed on a thermometer, indicating how well they attended to their voice or ignored a stranger's voice. fMRI, functional magnetic resonance imaging; MRI, magnetic resonance imaging; STG, superior temporal gyrus.

Figure 2

Multimodal neuroimaging features were associated with AH score change. (A) Greater AH reduction was associated with higher baseline AH. Eleven items on the PSYRATS auditory hallucination subscale were used to score AH. (B) Greater AH reduction was associated with greater right STG thickness (p < 0.05, uncorrected). The left panel shows the correlation between cortical thickness at each vertex and AH change. (C) Greater AH reduction was associated with lower dCB integrity (p < 0.05, corrected). The left panel shows the correlation between FA at each vertex and AH change (q_FDR < 0.05). (D) Greater AH reduction was associated with stronger DMN–PCC resting-state functional connectivity and stronger DMN–DLPFC anticorrelation (i.e., negative connectivity) (p < 0.05, uncorrected). The left panel shows the correlation between resting-state functional connectivity with individualized DMN seed at each voxel and AH change. AH, auditory hallucination; dCB, dorsal cingulum bundle; DLPFC, dorsolateral prefrontal cortex; DMN, default mode network; FA, fractional anisotropy; PCC, posterior cingulate cortex; PSYRATS, psychotic symptom rating scales; STG, superior temporal gyrus.