Background: Amygdala hyper-responsiveness to negative socio-affective stimuli have typically been demonstrated in patients with social anxiety disorder (SAD). Relative to conventional methods, there is emerging evidence that brain signal variability could be a better predictor of behavior than mean neural response.

Methods: We recruited 46 patients with SAD (mean age 31, 63% females) and 40 matched healthy controls (HC) to undergo 3 Tesla functional magnetic resonance imaging (fMRI) at 2 time-points, totaling 172 MRIsessions. Blood-oxygen level-dependent (BOLD-fMRI) was performed while viewing happy and fearful faces in blocks of 80 seconds. BOLD-fMRI data was reviewed by manually classifying signal from noise. Variability was calculated as each voxel’s standard deviation on signal across scanning-time. Multivariate partial least squares (PLS) estimated patterns of variability that separates patient from controls.

Results: PLS found one significant latent variable with cross-block covariance on 64%, permutated (x 1000) P<0.001, bootstrapped 95% confidence intervals on each condition, demonstrating less signal variability to happy faces in patients, relative to controls. This pattern of response was spatially located in several regions across the whole-brain, with large clusters appearing in bilateral amygdala, medial prefrontal cortex and posterior cingulate cortex/precuneus.

Conclusions: We found that neural response variability to positive socio-affective stimuli accurately separated patients from controls. It is likely that less signal variability highlights a deficit in effective emotion processing. We add to the growing literature on healthy individuals suggesting that task-specific brain signal variability contains useful information. The brain signal variability approach opens new avenues to evaluate and better understand brain function in common psychopathology.