Project 2: Discussion
In the present study, I examined whether individualized networks exhibit meaningful activity during emotion processing that is not captured by standard region-level (ROIs) approaches typically used to examine neural responses during tasks. Our aim was to extend the characterization of activity during a canonical emotion processing task beyond the amygdala, which is the current task target. To compare an individualized network approach to standard boundary methods, I examined activity within a set of meta-analytic regions of interest, as well as within networks defined by a group-averaged parcellation. By examining activity in meta-analytic ROIs alongside the individualized networks, I were able to evaluate the difference in observed activity at a region-level, as compared to a network-level. Furthermore, by also including a group-averaged parcellation, I were able to address differences between standard approaches for defining network boundaries (ie. using group-averaged data) versus precision neuroscience approaches that account for individual differences in neuroanatomy.
Emotion Processing Activation.
Amygdala — Although the amygdala exhibited significant activity in the Faces > Shapes contrast, looking at the faces and shapes conditions separately reveals that this effect is due to a slight increase above baseline in the faces condition and a slight decrease below baseline in the shapes condition. The lack of a robust amygdala activation in response to emotional faces was surprising, given that this task has been shown to elicit robust amygdala activation in adults (Fusar-Poli et al., 2009; Sabatinelli et al., 2011). However, a recent study examining a similar emotion processing task in a longitudinal developmental sample also found limited amygdala response (Flournoy, et al., 2023). Thus, it is possible that there are developmental differences in amygdala reactivity to this task, which warrants further investigation.
Meta-Analytic Regions of Interest — Two regions of interest (ROIs) exhibited significantly higher activity when processing emotional faces relative to shapes: left right fusiform. This finding is consistent with the literature, as left and right fusiform are known to have face-selective properties (Kanwisher, 1997). The right fusiform exhibited notably more activity than the left, which is also consistent with the widely established hemispheric lateralization of face processing in the fusiform (Meadows 1974; Kanwisher, 1997; Freiwald 2016). Even though increased activity in amygdala and the fusiform is in line with previous studies that adopted the same task (Somerville et al., 2018), I did not observe meaningful activity within other meta-analytically defined ROIs in higher order regions such as the medial prefrontal cortex (mPFC) or the middle frontal gyrus (MFG). Note that recent studies in two different developmental samples also did not observe robust activation in higher order regions (Somerville et al., 2018; Flournoy et al. 2023). While these higher order regions are often implicated in emotion processing tasks, the present task is at its core a simple visual matching task that can be completed without engaging higher order information processing. Regions derived from meta-analytic approaches inherently blurr differences in neural responses across task paradigms, making it challenging to draw global conclusions about emotion processing.
Group Networks — Of the 17 group networks examined, the only one that exhibited significant activity during emotion processing was the visual-central network (VisCent). We did not see significant activity within the dorsal attention network A of the group networks, which is the topologically most similar network to the individualized dorsal attention network B — (note the difference in naming convention B in individualized vs A in group). Examining activity within the group-averaged dorsal attention network A during each of the conditions alone (ie. Faces > Baseline and Shapes > Baseline), revealed significant activity during each condition, however when this activity was contrasted (Faces > Shapes) there was no meaningful difference in activity between the conditions.
Individualized Networks — Of the 15 individualized networks examined, two exhibited significant activity during emotion processing: the visual-central network (VIS-C) and dorsal attention network B (dATN-B). The visual-central network exhibited the highest level of activity during emotion processing when compared to the other networks in the individualized parcellations. Given the visual-central network’s role in processing information in the central visual field, this activity likely reflects the additional visual processing elicited when viewing complex emotional faces relative to basic geometric shapes (Levy, 2001). The minimal activity in the visual-central network during the shape condition (three white circles on a black background) compared to the baseline (white plus sign in the center of black screen) is in line with this interpretation. The only other individualized network that exhibited significant activity during the Face > Shapes contrast was the dorsal attention network B (dATN-B). The canonical dorsal attention system, which roughly comprises attention network A, attention network B and PM-Ppr of the individualized networks, was one of the first networks to be identified using functional connectivity approaches (Fox et al., 2008) and has been broadly implicated in visual and attention tasks (Corbetta & Shulman, 2002; Vincent et al., 2007; Patel et al., 2015). Recent work has identified that this large-scale network segregates into at least two dissociable networks (dorsal attention network A and B), where dorsal attention network B couples with retinotopic visual regions, while dorsal attention network A doesn’t (Braga et al., 2017). These dissociable patterns suggest that dorsal attention network B may support stimulus-driven (bottom-up) control of attention, while dorsal attention network A may be more involved in goal-directed (top-down) attention. Our results corroborate these findings whereby I see dorsal attention network B activation in this very simple visual matching task, but do not see dorsal attention A. Taken together, these results suggest that the canonical emotion processing task used in the present study primarily engages networks associated with visual processing and stimulus driven attention.
Individualized vs Group-Averaged Approaches. While significant activity was observed in the visual-central network across both individualized and group averaged approaches, the individualized visual-central network (VIS-C) elicited a notably stronger response than the group-average equivalent (VisCent). Significant activity was also observed in dorsal attention network B (dATN-B) of the individualized networks, but not the equivalent group average network (DorsAttnA). As noted, the group average dorsal attention network A did exhibit significant activity during both the faces and shapes conditions, however the difference in activity between the conditions was negligible. This suggests that the group-averaged networks, while exhibiting similar general tendencies to individually defined networks, are not sensitive enough to capture meaningful functional differences to the degree that individualized networks can. These findings provide an additional example of individualized networks being better aligned to task-evoked activity and therefore providing a better framework for detailing functional activation in response to tasks (Chong et al., 2017; Tavor et al., 2016; DiNicola et al., 2020; Salvo et al., 2021).
Beyond ROIs - Confirming Network-Level Activation. Significant activity during emotion processing was observed at both a regional level (amygdala, left and right fusiform), as well as at a network-level (individualized visual-central network and dorsal attention network B). Importantly, the degree of activity in the right fusiform and the individualized visual-central network (VIS-C) was comparable. While the fusiform was the only cortical region that showed meaningful activity to faces relative to shapes, significant activation for this contrast was observed both within the visual-central network, as well as individualized network dorsal attention B. The fusiform primarily falls within the visual-central network, but also exhibits a slight overlap with dorsal attention B. Given this overlap it is possible that the observed activity within the visual-central network and dorsal attention network B is just driven by activation within the fusiform. The networks are each composed of thousands of vertices that each exhibit a level of activity. By examining the distribution of vertex-level activity within each network I can assess whether a subset of vertices are driving up the average activation observed within the network, or whether the vertices across the whole network are showing increased activation in response to task. A visual assessment of session-level emotion processing vertex-level activity illustrated that when a network shows significant activity in this task, the majority of the vertices that comprise the given network demonstrate a collective increase in activity. This visual observation was quantified by calculating the percent of vertices that exhibit activity greater than zero. The visual-central network showed the highest percent of vertex activity above zero for faces relative to shapes, followed by dorsal attention network B. We additionally found strong correspondence between the individual network boundaries and the pattern of activity in the faces > shapes contrast, confirming previous results (Chong et al., 2017; Tavor et al., 2016). Finally, I additionally observed that the vertex-level activity within a network can be meaningfully distinguished from other networks, suggesting that all networks exhibit unique levels of activity in this task that are quantifiable at the vertex-level. These findings imply the individualized network serves to explain the global response profile of activity across the whole brain, not just within regions that exhibit significant activity. In sum, these findings suggest the presence of network-level activity in this emotion processing task.
Conclusion. In the present study, I found evidence of network-level activity in response to a canonical emotion processing task in two networks involved in visual information processing and attention (the visual-central network and dorsal attention network B). Though expected region-level activity was observed within the amygdala and the fusiform, the individualized network-level approach allowed for a more comprehensive characterization of task-evoked activity. Our findings suggest individualized network-level approaches can be used in conjunction with region-level analysis to produce a rich multi-level account of functional activation in response to task. Such approaches provide a task independent global framework that can be utilized across paradigms in a way that current aggregation approaches (ie. meta-analyses) cannot. In sum, future studies aiming to utilize this canonical task, and others like it, may benefit from including an individualized network-level approach in their analysis.