Beyond regional specialization — using individualized networks to characterize cortical brain function during emotion processing
The identification of cortical networks in the human brain using functional connectivity approaches revolutionized the field of cognitive neuroscience. Since the initial discovery that intrinsic functional activity in the human brain exhibits reliable large-scale network organization, thousands of studies using various methods have explored inherent brain organization. The rise of functional connectivity marked a shift from traditional fMRI techniques, which rely on controlled task paradigms to isolate and study functions within small specialized brain regions. While region-level approaches have led to a wealth of knowledge about the neural substrates of specific phenomena that can be isolated in this way, such approaches impede the characterization of brain function within a systematic global framework. For this reason, the advent of functional connectivity and discovery of the cortical networks has been groundbreaking, as it has provided an unparalleled opportunity to examine brain-wide functional organization in humans. In the present project, I conducted three projects aimed at extending our understanding of the cortical brain networks and their role in human brain organization and function.
Through the use of functional connectivity analyses, the field has established numerous functional brain atlases that map the topography of the cortical networks. These network parcellation schemes are used as a priori boundaries to examine individual differences in network organization and function. Researchers typically select a single parcellation from the many available for use in their work. This practice has become ubiquitous, particularly in individual difference studies interested in linking variability in functional connectivity to other characteristics such as age and psychopathology, yet hinges on a series of fundamental assumptions that have not previously been evaluated. I did so in project 1. The primary finding from project 1 was that parcellation selection meaningfully impacts the results of studies of individual differences in network organization and function. I found that associations between within-network functional connectivity and factors such as poverty, age, and cognitive function (i.e., inhibitory control) varied significantly based on which parcellation was used to identify the network of interest. These findings challenge the validity of studies that rely on group-averaged a priori parcellations for network identification and highlight the potential need for alternative approaches to identifying cortical networks that do not depend on these fundamental assumptions.
Precision neuroscience approaches have moved away from traditional group-averaging methods used to create parcellations that also obscure individual variability in brain anatomy and function. While group-averaged data helped establish foundational frameworks for understanding brain organization, it has been shown to oversimplify and mischaracterize functional organization. By focusing on deeply sampled individuals, precision neuroscience has revealed that cortical networks are highly individualized, often comprising multiple distinct networks with dissociable functional roles that group-averaging fails to capture. This approach has provided a more accurate and nuanced understanding of brain function, highlighting the importance of individualized analysis in neuroscience research.
In project 2, I explored whether individualized networks could reveal meaningful activity during a canonical emotion processing task that standard region-level approaches (ROIs) did not capture. The task explored is one that has been included in some of the largest data-collection efforts in the field of cognitive neuroscience, yet has been plagued with issues of poor reliability, particularly within the amygdala, the region targeted in this task. The primary aim of project 2 was to examine whether activity in this canonical task can be meaningfully analyzed at a network-level using individualized networks, offering an alternative approach for activity characterization. By comparing activity within meta-analytic ROIs, group-averaged networks, and individualized networks, we found that individualized network-level analysis revealed significant task-related activity in networks associated with visual processing and stimulus driven attention, beyond the expected regional activity within the amygdala and fusiform. These results highlight that incorporating individualized network-level approaches alongside traditional region-level analysis offers a more comprehensive understanding of neural responses during tasks that overcomes the limitations of standard aggregation methods. This suggests that future research may benefit from integrating individualized network analyses to enrich the characterization of functional brain activity during tasks.
Following the observation that we do indeed see meaningful network-level activity in a canonical emotion processing task, in project 3, I investigated whether activity within the individualized networks improved test-retest reliability, beyond that observed within the amygdala, the original task target. The hope was that activity within individualized networks might be more reliable than activity derived from smaller ROIs, particularly the amygdala. If so, this would provide an opportunity to continue to explore factors that underpin individual differences in emotion processing using this canonical task. I compared the reliability of activity within individualized networks to activity within the amygdala, meta-analytic ROIs, and group-averaged networks. Additionally, I examined whether network-level activation was linked to stress, depression, and anxiety symptoms, both between and within individuals, as well as association with task performance. Ultimately, the individualized network approach did not improve test-retest reliability. Although activity within the individualized networks was associated with task performance, it was not associated with stress or psychopathology, either between or within individuals. The findings from project 3 add to the growing evidence calling into question the value of this highly utilized canonical emotion processing task for studying individual differences.
In sum, through the three projects detailed here I found that group-average approaches to identifying network level organization do not produce reliable connectivity results and produce diminished effects relative to individualized approaches in studies of task-evoked activity. Individualized networks can be used with existing tasks in conjunction with standard region-level techniques to establish multi-level characterizations of task-evoked activity. And finally, caution is warranted when considering the use of this canonical emotion processing task. Our findings add to the growing evidence that this task is not well positioned to draw meaningful conclusions about individual differences in emotion processing, given the poor test-retest reliability observed across different levels of analysis,in regions and networks alike. These findings are important given that this task is being used in some of the largest data collection efforts in the field with the goal of understanding individual differences in brain function.