IMVEST, an immersive multimodal virtual environment stress test for humans that adjusts challenge to individual's performance

Figures

• 

  Fig. 1. Description of IMVEST main elements. (a) Schematic for the logic governing IMVEST (consult Appendix 1-Fig. S1 for a more detailed flowchart). Settings are loaded from a text file and a room of 5 by 3 m is generated with the Unity® game engine. A 2.5 min training period precedes the 10 min test. During training, stressful elements are turned OFF, difficulty is not adjusted and intervals between arithmetic tasks last exactly 1.75 s without jitter. For the test, all these elements are turned ON. The test contains arithmetic tasks that are presented on screen during a certain amount of time (display time). During this time, the participant can respond pressing the controller trigger if the equality is true, or the controller button if the equality is false. Failing to respond within the display time is taken as an incorrect response. Incorrect responses trigger the explosion of a random tile in the room, leaving an open gap on the virtual floor. After each arithmetic task, the two performance variables, an overall measure of performance and a measure of rolling performance of the las 5 trials, are updated. The latter is titrated to be bound between 40% and 60% by incrementing or…

• 

  Fig. 2. Experimental protocol applied to the IMVEST's validation and characterization study and main physiological outcomes. (a) Experimental protocol applied to each participant, including the different experimental phases and saliva sampling periods for cortisol analysis (s1-s4). Representative durations for sensor placement. Cortisol results corrected for circadian influence: (b) across the experiment; and (c) area under the curve with respect to increases (AUCi). (d) Schematic depiction of sensor placement and VR apparatus. Stress effects in ANS variables related to the sympathetic response (e, f, g) and to the parasympathetic response (h, i, j). Alpha significance levels: *p < 0.05, **p < 0.01, ***p < 0.001. P values corrected for the multiple comparisons across blocks. Data samples are represented by their mean (dot) and its standard error (bars).

• 

  Fig. 3. Characterization of the behavioral changes during participants' IMVEST exposure and their association with physiological variables. (a) Selected behavioral features and computations of ‘difficulty adjustments’ across blocks T1-T4. Data samples are represented by their mean (dot) and its standard error (bars). (b) Correlations between behaviors and key physiological markers. Spearman correlation coefficients between the averaged variables across T1 to T4 are shown in the upper triangular half of the matrix; repeated-measures correlation (rm-correlation) throughout T1 to T4 are shown in the lower triangular half of the matrix. Effects between physiology and behavior are highlighted in yellow. Alpha significance levels: *p < 0.05, **p < 0.01, ***p < 0.001. P values are corrected for the multiple comparisons between variables for each correlation metric. (c) Position heatmap for each block in the 5 by 3 tiles virtual room of IMVEST. (d) Position preference for different parts of the virtual room: center (3 central tiles between 1 and 4 m in the z coordinate and 1 and 2 m in the x coordinate); tiles at the corner of the room; edges (all tiles except central ones). Data samples are…

• 

  Fig. 4. Characterization of the autonomic response to IMVEST and development of a PLS model. (a) Statistic heatmap for independent samples t-test (stress vs. control) for each ANS variable (see Table 2 for variable description). Variables are grouped together with hierarchical clustering according to the similarities in the t-test test statistics to facilitate interpretation. A clear distinction between stress and control can be observed at T1, T3 and T4 indicating parasympathetic withdrawal and sympathetic activation, with more than 50% of the tested variables showing significant differences at T3 and T4. Alpha significance levels: *p < 0.05, **p < 0.01, ***p < 0.001. P values corrected for multiple comparisons between variables in each block. (b) Statistical validation of each individual PLS model by non-parametrically testing the significance of their area under the curve (AUC). Individual models' AUC are compared to the AUC distribution of 10 k models trained on datasets with randomly shuffled labels at the participant level. (c) Average of the PLS coefficient scores across all individual models, informing about the magnitude and direction of the influence that each variable exerts…