Fig 1.
Study Day 2 experimental tasks: Order, number of repetitions, duration, and stimuli.
Tasks are colored by role. Gray depicts task training and application of psychophysiology recording apparatus. Blue depicts structural image acquisition. Orange depicts functional image acquisition. Identification and Modulation blocks of the fMRI acquisition summarize the relevant trial types used within that task (see Neuroimaging section for abbreviations). *Training of real-time multivariate pattern analysis predictive models was performed concurrently with the Resting State task of the fMRI acquisition.
Fig 2.
Summary of experimental task trial designs.
(Id-PS): Identification task passive stimulus trials, which were identical to Modulation task passive stimulus (Mod-PS) trials. (Id-CR): Identification task cued-recall trials. (Mod-FS): Modulation task feedback-triggered stimulus trials. (Bottom): Depiction of a hypothetical Mod-FS trial for the experimental design. The dashed line represents the trigger threshold and bounds the hyperplane distance at which the cue stimulus will be triggered by the real-time valence estimate as a function of time. As depicted, this threshold decreases linearly to zero commencing at 20 s of feedback. This trial type is of fixed length; therefore, the ITI duration is a function of the time required to trigger the stimulus via feedback. If the real-time valence estimate does not surpass the trigger threshold prior to the threshold reaching zero then the stimulus is triggered by default, denoted “Emergency trigger”, followed by the minimum ITI.
Fig 3.
Normative valence and arousal scores for stimuli selected for each of the four experimental trial types.
Summary statistics for Identification task stimuli are as follows: Id-PS valence [mean (std. dev)] 5.04 (1.95); Id-PS arousal [mean (std. dev)] 4.95 (1.40); Id-CR valence [mean (std. dev)] 5.30 (1.95); Id-CR arousal [mean (std. dev)] 4.99 (1.51). There were no significant differences in affect properties between the Id-PS and Id-CR cue stimuli for either valence (p = .49; signed rank; α = .05; h0: μ1 = μ2) or arousal (p = .86; rank-sum; α = .05; h0: μ1 = μ2). Summary statistics for the Modulation task stimuli are as follows. Mod-PS (pos. valence cluster) valence [mean (std. dev)] 7.41 (.30); Mod-PS (neg. valence cluster) valence [mean (std. dev)] 2.08 (.36); Mod-FS (pos. valence cluster) valence [mean (std. dev)] 7.35 (0.32); Mod-FS (neg. valence cluster) valence [mean (std. dev)] 2.03 (0.41). Between the Mod-PS and Mod-FS stimuli in the positive valence cluster, there were no significant differences in valence (p = .60; rank-sum; α = .05; h0: μ1 = μ2) nor arousal (p = .25; rank-sum; α = .05; h0: μ1 = μ2). There were also no significant group differences in affect properties between the Mod-PS and Mod-FS stimuli in the negative valence cluster, either for valence (p = .74; rank-sum; α = .05; h0: μ1 = μ2) or arousal (p = .54; rank-sum; α = .05; h0: μ1 = μ2).
Table 1.
Multivariate neural decoding performance.
Fig 4.
Group-level encodings of affective state processing.
Color gradations indicate the group-level t-scores of the encoding parameters (red indicating positive valence or high arousal, blue indicating negative valence or low arousal). T-scores are presented only for those voxels in which encoding parameters survived global permutation testing (p < .05). Image slices are presented in MNI coordinate space and neurological convention. Maximum voxel intensity is |t| = 6.0, i.e., color saturates for t-scores with absolute values falling above this value.
Fig 5.
Distribution of average feedback scores at the moment of FT-PO trial stimulus trigger.
Fig 6.
Effects of volitional self-induction of positive valence on affect processing bias of subsequent image stimuli.
The figure graphically depicts the effect sizes estimated for the primary experimental manipulation, i.e. the self-induction trial type (feedback-triggered stimulus, Mod-FS, versus passive stimulus, Mod-PS), denoted self-induction, on the decoded valence processing of the subsequent image stimulus while controlling for the effects of the normative valence score of the stimulus, the decoded valence processing of the previous fMRI volume, normative arousal score of the image stimulus as well as subjects’ age and sex and the two-way interactions between self-induction trial type and the normative valence score of the stimulus as well as the decoded valence of the previous fMRI volume. Statistically significant effects are colored blue (positive effects) or red (negative effects). Non-significant effects are colored gray. *The effect size of the decoded valence processing (β = .802) of the previous fMRI volume was omitted from the figure to elevate the contrast between the smaller effect sizes.
Fig 7.
Estimation and validation of explicit intrinsic affect regulation effects within the cued-recall task.
The figure depicts the effect size of cue affect processing in explaining affect processing occurring during recall (controlling for time lag in the 4 repeated measures of recall per each measure of cue). Here affect processing measurements are Platt-scaled hyperplane distance predictions, Pr(∙), of our fitted support vector machine models. Valence and arousal dimensions of affect are predicted by separate models. The figure’s scatterplots depict the group-level effects computed using linear mixed-effects models which model random effects subject-wise. Bold red lines depict group-level fixed-effects of the cue affect. Bold gray lines depict significant subject-level effects whereas light gray lines depict subject-level effects that were not significant. The figure’s boxplots depict the group-level difference between each subject’s affect regulation measured during the cued-recall trials in comparison to surrogate affect regulation constructed from the resting state task. The bold red line depicts the group median difference in effect size between task and surrogate. The red box depicts the 25-75th percentiles of effect size difference.