Multivariate pattern classification and representational similarity analysis : an application of real-time functional magnetic resonance imaging-based neurofeedback decoding in major depressive disorder

RWTH Publications (RWTH Aachen)

Abstract

Neurofeedback (NF)-supported cognitive reappraisal training has shown promise for improving emotion regulation in psychiatric disorders, particularly major depressive disorder (MDD). Previous findings from a real-time functional magnetic resonance imaging NF (rt-fMRI NF) study indicated that left compared to right ventrolateral prefrontal cortex (vlPFC) modulation was associated with faster learning and greater emotion regulation, consistent with established models of cognitive reappraisal lateralization (Keller et al., 2021). Despite these promising findings, traditional univariate whole-brain analyses failed to reveal significant neural evidence for such lateralized effects or corresponding group differences between MDD patients and healthy controls. Using multivariate pattern analysis (MVPA), this study examined rt-fMRI NF data from 37 patients with MDD and 37 healthy controls to investigate neural representations of lateralized NF training. A whole-brain searchlight multivariate pattern classification (MVPC) decoded brain activity distinguishing left- and right-NF conditions and revealed high classification accuracies in areas involved in emotion regulation such as supplementary motor area (SMA), inferior frontal gyrus (IFG), insula, and anterior cingulate cortex (ACC); these effects even survived very conservative correction for multiple comparisons (pFWE < 5*10-11). In addition, representational similarity analysis (RSA)-based pattern consistency measured neural encoding stability in 20 regions of interest (ROIs) related to emotion regulation and showed significantly lower values in the MDD compared to the control group during the left-NF condition in the bilateral precentral gyri, the right IFG (opercular part), and the left SMA (p < .05; no group difference during right-NF condition, all p > .1). These results demonstrate that multivariate approaches can capture subtle, distributed neural differences in NF-guided emotion regulation, which may be overlooked by conventional univariate analyses. Significant differences in the searchlight analysis suggest that the laterality of the NF target is associated with local network reconfiguration and such connectivity modulation is critical for cognitive flexibility and executive function. Further, the encoding deficits in MDD during left-targeted NF may reflect a neural mechanism underlying its pathophysiology. MVPA offers a valuable tool for characterizing complex cognitive and affective processes such as training of emotion regulation, with implications for understanding dysregulated cognitive processes in MDD and its neuroplastic treatment.

Authors 0

1. Author list not loaded yet.