Functional connectivity analyses for task-based fMRI data are generally preceded by methods for identification of network nodes. As there is no general canonical approach to identifying network nodes, different identification techniques may exert different effects on inferences drawn regarding functional network properties. Here, we compared the impact of two different node identification techniques on estimates of local node importance (based on Degree Centrality, DC) in two working memory domains: verbal and visual. The two techniques compared were the commonly used Activation Likelihood Estimate (ALE) technique (with node locations based on data aggregation), against a hybrid technique, Experimentally Derived Estimation (EDE). In the latter, ALE was first used to isolate regions of interest; then participant-specific nodes were identified based on individual-participant local maxima. Time series were extracted at each node for each dataset and subsequently used in functional connectivity analysis to: (1) assess the impact of choice of technique on estimates of DC, and (2) assess the difference between the techniques in the ranking of nodes (based on DC) in the networks they produced. In both domains, we found a significant Technique by Node interaction, signifying that the two techniques yielded networks with different DC estimates. Moreover, for the majority of participants, node rankings were uncorrelated between the two techniques (85% for the verbal working memory task and 92% for the visual working memory task). The latter effect is direct evidence that the identification techniques produced different rankings at the level of individual participants. These results indicate that node choice in task-based fMRI data exerts downstream effects that will impact interpretation and reverse inference regarding brain function.
Keywords: Activation likelihood estimates; Degree centrality; Individual differences; Node identification.