Individuals diagnosed with social anxiety disorder were asked to participate in the VR therapy program developed and verified by the authors [25]. Two psychiatrists and two licensed psychologists performed collaborative work to develop a social-anxiety scenario (during a team meeting in class), and the Art & Technology Lab at the Korean National University of Arts designed VR environments using the Unity game engine (Unity Technologies). Motion capture was conducted using Sensor Suite (Rokoko Electronics) for VR character animation, and voice acting was performed by 4 actors from Korean National University of Arts drama school. In our previous work [25], we proved that our participatory and interactive VR intervention had significant beneficial effects on depression, anxiety, shame, rumination, and social phobia.

The participatory VR intervention for social anxiety symptoms consisted of 3 stages (introduction, core, and finishing) and was divided into 3 levels of difficulty (easy, medium, and difficult). All participants used a VIVE (HTC Corporation) VR headset, and the participants’ heart rates, skin galvanic response, and eye movements were measured during the VR experience.

In the introduction stage, the participants were invited to choose their avatar to learn how to use VR, calm their minds, and help them adapt to VR during the meditation-based warm-up. A voice guide was provided to help the participants relax and breathe while observing trees gently shaking to calm their minds. Through this, the participants not only adapted to the VR system but also moved to the next stage in a stable state of mind. The introduction stage was configured to take approximately 5 minutes.

The core stage was designed to provide a solution to individuals’ exposure and participation in social anxiety situations by providing them with a VR environment in which a college student group was meeting for the first time to discuss an assigned task and introduce themselves. In the virtual setting, 7 to 8 nonplayer characters introduced themselves then the participants with social anxiety disorder were asked to introduce themselves. At the easy level, each participant took turns introducing themselves in a normal and calm manner, starting with the nonplayer characters. In the difficult situation of encountering an unfavorable reaction from nonplayer characters, the nonplayer characters became increasingly distracted while listening to the participant’s self-introduction, making more small talk among themselves, concentrating harder on other tasks, or staring intently at each other. The medium level consisted of distractions between the easy and difficult levels. The core stage was configured to take approximately 7 to 8 minutes.

In the finishing stage, the participant, as in the introduction stage, once again experienced meditation-based VR that calmed the mind while observing a gently shaking tree and controlling their breathing. The VR program ended by providing general cognitive and behavioral psychoeducation for social anxiety disorder in both voice and text form via the VR system. The final stage was configured to take approximately 3 minutes.

All participants were asked to complete a total of 6 VR sessions, each consisting of the 3 stages. In one visit, each participant could complete up to 2 VR sessions as long as the participant took at least a 2-hour break between sessions. All participants started at the easy VR level. From the second session onward, the participants were asked to select the level of difficulty they desired. The difficulty level was increased, maintained, or decreased at the participant's request to provide an individual-tailored intervention. The researchers stayed with the participants throughout the VR sessions to address emergencies such as extreme anxiety or panic attacks.

All participants underwent fMRI while performing a self-referential processing task and completing self-report questionnaires, including the Korean version of the Social Phobia Scale (SPS) (H. Kim, unpublished) and Post-Event Rumination Scale (PERS) [28]. The participants with social anxiety disorder underwent fMRI and assessments before and after treatment, while the control participants underwent fMRI only once. The recommended sample size for a task fMRI is 20; our sample size (n=25 in the social anxiety disorder group at baseline, n=21 in the social anxiety disorder group at follow-up) seemed to have adequate statistical power [29]. The study was approved by the institutional review board of Korea University Anam Hospital in accordance with the Declaration of Helsinki. This study was registered (Clinical Research Information Service KCT0003854).

A revised version of the Personal Relevance Rating Task (PRRT) was used for the fMRI task [30]. The task consisted of 2 runs with a duration of 9 minutes 22 seconds per run, and each run included 40 trials (yielding a total of 80 trials). Each stimulus word was projected onto an angled mirror mounted on the head coil for 2 seconds using E-prime software (Psychology Software Tools). Between experimental stimuli, a mask (row of X's; 10.8 seconds) and a fixation cue (1 second; row of X's with prongs around the center X) were presented. The order of presentation of all stimuli was counterbalanced.

To build the set of stimuli, we selected 10 positive words, 10 negative words, and 20 neutral words for the experiment, balanced for arousal level, emotional valence, and word length. In addition to the normed emotional adjectives, 10 positive words and 10 negative words generated by the participants were added to the word list for the fMRI PRRT task. The instructions were as follows: “Generate ten positive words that best represent your strengths” and “Generate ten negative words that best represent your weaknesses.” All of these words in both normed stimuli (10 positive, 10 negative, 20 neutral words) and participant-generated stimuli (10 positive, 10 negative words) were selected from the Korean emotion words list [31].

During the fMRI, participants were instructed to make a judgment on each word as soon as possible by pressing the buttons for “not relevant to me,” “somewhat relevant to me,” or “relevant to me” after the word projected. Prior to the scan, participants completed a practice session to ensure that they understood the task. All responses and reaction times were recorded.

Neuroimaging was performed using an MRI machine (3T Siemens Tim trio) equipped with a 12-channel head coil at the Brain Imaging Center, Korea University. Axial T2*-weighted images (echo time 26 ms, repetition time 2000 ms, flip angle 80°, field of view 210 mm, voxel size 2.5×2.5×3.4, slice thickness 3.4 mm, matrix size 84×84). A single functional run consisted of 278 volumes with 37 sequential axial slices each. In addition, structural T1-weighted images (208 slices; echo time 1.89 ms, repetition time 1670 ms, flip angle 9°, field of view 250 mm, matrix size 256×256) were also obtained to aid with spatial normalization.

Anatomic T1 and functional T2* images were analyzed with Statistical Parametric Mapping software (version 12; Welcome Department of Cognitive Neurology). The echo-planar images were corrected for slice acquisition time and realigned to correct for rigid body transformation, then the individual's anatomic image was coregistered to the mean functional image. The echo-planar images were subsequently normalized using nonlinear transformation parameters obtained by registering individual T1-weighted images to the Montreal Neurological Institute template [32] and smoothed with an isotropic 6-mm³ full-width-at-half-maximum Gaussian kernel.

Demographics did not differ significantly between the 2 groups (sex, P=.71; age, P=.36; handedness, P=.51; education, P=.19), but compared with the controls, individuals with social anxiety disorder had higher SPS (Mann–Whitney U test: Z=−5.67, P<.001, r=−.83) and negative PERS (F_1,47=79.92, P<.001, η²_p=.64) scores as well as lower positive PERS scores (Mann–Whitney U test: Z=4.28, P<.001, r=−.63) (Table 1).

Statistically significant multivariate effects were found between the social anxiety disorder and control groups for the response ratings and reaction times for words of every valence (positive, neutral, and negative) (Wilks lambda=.58, F_3,43=10.41, P<.001, η²_p=.42). Participants who were highly socially anxious rated positive words as less (F_1,45=14.23, P<.001, η²_p=.24) and negative words as more personally relevant (F_1,45=22.53, P<.001, η²_p=.33) than controls did. There were no significant group differences in reaction time for any valence (Wilks lambda=.93, F_3,43=1.11, P=.36, η²_p=.07) (Table 1).

In the whole-brain analysis, compared with the controls, the social anxiety disorder group showed increased activation of the bilateral inferior frontal gyrus and left middle temporal gyrus during positive self-referential processing (positive-word images>neutral-word images). In individuals with social anxiety disorder, there was also increased left insula activation and decreased right middle frontal gyrus activation in response to negative self-referential processing (negative-word images>neutral-word images) (Figure 1 and Table 2).

Repeated measures ANOVA for the response ratings and reaction times in the social anxiety disorder group (n=21) showed marginally significant changes in the ratings for the positive words (F_1,19=3.85, P=.06, η²_p=.16) and negative words (F_1,19=3.77, P=.07, η²_p=.16). No statistically significant changes were found for reaction times before and after VR treatment (Table 3).

The SPS and negative PERS scores of the social anxiety disorder group decreased remarkably (F_1,20=9.83, P=.005, η²_p=.33; Wilcoxon signed-rank test: Z=–3.32, P<.001, r=.51, respectively) following VR therapy. There were no significant changes in positive PERS scores between baseline and follow-up (P=.75) (Table 3).

The social anxiety disorder group had significantly increased activation of positive self-referential stimuli (positive>neutral) in the right posterior cingulate cortex/precuneus, lingual gyrus, left inferior temporal gyrus, precentral gyrus, and postcentral gyrus after treatment. Moreover, increased activation was found during negative self-referential processing (negative>neutral) in the left middle occipital gyrus, parahippocampus, left Rolandic operculum, left superior frontal gyrus, and left caudate nucleus (Figure 2, Table 4) at follow-up compared with that at baseline.

Changes in lingual gyrus activation associated with positive self-referential processing were related to alleviated social anxiety (SPS, Spearman ρ=–.52, P=.02) and decreased rumination on the negative events (negative PERS, Spearman ρ=–.61, P=.005) in the postintervention sessions. In addition, mean percentage BOLD signal changes in regions revealing significant changes following the VR intervention were not correlated with either positive PERS score at follow-up or the other symptom change rates in social anxiety disorder.

Another overarching aim of this fMRI study was to identify neural changes associated with self-referential processing as well as anxiety symptom reduction in social anxiety disorder following VR therapy designed to teach self-assertiveness. Because the control group did not participate in the therapy sessions, it cannot be decisively stated that the fMRI data obtained were the direct results of the VR therapy intervention. What is clear, however, is that there were significant changes in brain activity during positive self-referencing processing in the social anxiety disorder group who completed the VR sessions. That is, the participants showed not only significant reductions in social anxiety disorder symptoms but also neural changes in the right posterior cingulate cortex/precuneus, lingual gyrus, left inferior temporal gyrus, postcentral gyrus, and right precentral gyrus during positive self-referential processing after VR therapy.

We found increased activity in the posterior cingulate cortex/precuneus of individuals with social anxiety disorder. These cortical regions, along with the precentral gyrus, exhibit enhanced activation during the processing of positive self-referential stimuli in healthy volunteers [30]. The posterior cingulate cortex/precuneus also plays a crucial role in integrating autobiographical information regarding the self [47]. Interestingly, a recent finding from experimental research suggested that social anxiety relates to impaired memory for social scenarios that ended with positive outcomes [48]. Altogether, our results may reflect increased cognitive efforts to access autobiographical memory while processing positive self-referential information in social anxiety disorder. This result needs to be further studied using an autobiographical memory task and comparison with a control group.

After VR therapy, the activation of the lingual gyrus and inferior temporal gyrus during positive self-referential processing also increased. According to previous studies [49,50] using healthy populations, stronger activations of the lingual gyrus, inferior temporal gyrus, and posterior cingulate cortex are closely related to self-referential processing. Intriguingly, the greater the change in lingual gyrus activation while processing positive self-referential information, the lower the tendency to ruminate on negative life events and the lower the social anxiety levels following the VR intervention session. Additionally, the inferior temporal gyrus has been reported to play an important role, along with the posterior cingulate cortex, in judging whether a series of stimuli is self-related [51]. Therefore, increased BOLD signals in these cortical regions may reflect the facilitative effect of VR therapy on the process by which the individuals with social anxiety disorder accept the positive-valence words as being relevant to them.

The precentral and postcentral gyri also showed increased activation during positive self-referential processing in individuals with social anxiety disorder upon completing the VR sessions. A similar brain activation pattern was observed in individuals with anxiety disorders who used cognitive reappraisal [52]. We conjecture that activation in the postcentral gyrus represent the therapeutic effects of the social anxiety–focused VR program. Further research is needed to investigate whether the therapeutic effect of our VR program could be extended to individuals with other clinical conditions that exhibit reduced postcentral gyrus activity, such as those with a generalized anxiety disorder [53] or a history of childhood abuse [54].

Participants with social anxiety disorder exhibited greater activation in cortical and subcortical regions, which are known to be involved in somatosensory integration, during the processing of negative self-referential stimuli after VR therapy than at baseline. Similar findings were reported from an fMRI study [19,55] showing the neural mechanisms of the cognitive reappraisal of negative self-beliefs in individuals diagnosed with social anxiety disorder; in particular, the participants with social anxiety disorder demonstrated greater cognitive and somatosensory brain responses while reappraising negative self-beliefs. The most recent study using healthy adults [56] showed that cognitive bias modification for interpretation resulted in significantly greater activations of the somatomotor and somatosensory areas and occipital lobe. Our results may suggest that the VR-based therapy facilitated the perceptual or sensory-motor processing of negative self-referential stimuli in individuals with social anxiety disorder who had been willing to avoid negative feedback cues. Further research using a control group is needed to explore whether such an enhancement of brain activation can be linked to an approach-oriented strategy with negative-valence stimuli in social anxiety disorder.