Can cognitive functions be inferred from neuroimaging data? A reverse inference meta-analysis of theory-of-mind tasks

Cognitive neuroscience research using functional magnetic resonance imaging (fMRI) has predominantly focused on localizing patterns of neural activity associated with human cognitive functions. This approach, known as forward inference, has been pivotal in pinpointing brain areas engaged during specific cognitive tasks and testing hypotheses about brain-behavior relationships. In contrast, the use of reasoning from brain activation to cognitive functions, known as reverse inference, has been considered more informative because it allows researchers to interpret neural activity patterns to make inferences about the cognitive domain likely at play. Crucially, reverse inference considers how selectively the area is activated by the cognitive function under investigation, which is particularly important given the multifunctional nature of many cortical and subcortical areas. Nevertheless, the practical application of reverse inference in fMRI research remains methodologically challenging. Here, we performed a meta-analytic reverse inference analysis of brain activations related to Theory-of-Mind (ToM) tasks to evaluate whether this approach can effectively identify selective brain areas recruited for this critical human cognitive function. Leveraging data from the BrainMap database, we analyzed 223 published fMRI experiments involving ToM tasks (1069 healthy participants and 1526 activation foci) and compared these findings to fMRI data from other tasks stored in the BrainMap database (110 distinct cognitive tasks, 8154 published experiments, 127112 healthy participants, and 66649 activation foci). To achieve this, we applied Bayes fACtor mOdeliNg, a novel Bayesian-based, data-driven, hypothesis-free method that provides posterior probability distributions for the evidence of selectivity with respect to a given mental process. We found that several brain areas commonly recruited in ToM tasks (e.g., bilateral inferior frontal gyri, superior temporal cortices, and posterior cingulate cortex) show a low level of selectivity (P < 50%), indicating their involvement across multiple cognitive domains. The results also revealed a small, organized set of highly selective areas (P > 90%; e.g., bilateral superior frontal gyri, inferior temporal gyri, right precuneus, and anterior cingulate cortex) that map the cognitive function of ToM. These results provide a more refined and nuanced approach to understanding the neural basis of cognition, offering valuable insights for the development of formal cognitive ontologies and the refinement of brain-cognition models.