Connectome-based predictive modeling of mind wandering within densely-sampled individuals

Poster Session F - Tuesday, April 16, 2024, 8:00 – 10:00 am EDT, Sheraton Hall ABC

Lotus Shareef-Trudeau¹ (ls3623@drexel.edu), Aaron Kucyi¹, Shao-Min Hung²; ¹Drexel University, ²Waseda Institute for Advanced Study, Waseda University

Neuroimaging studies have evidenced that mind wandering can be predicted from whole-brain patterns of functional connectivity at a group-level. Given the large amount of variability present in individual mind wandering experiences, whether group-level models are effective in predicting mind wandering at the individual level remains to be seen. Using data collected in a dense-sampling fMRI study, we sought to answer this question. Participants (n=3) performed a “resting state” visual fixation task with interspersed thought probes across multiple sessions, resulting in 10 hours of scanning and 350 probe trials each. We applied connectome-based predictive modeling (CPM) within individuals, based on functional connectivity within 30-second windows prior to thought probes, using five-fold and leave-one-trial-out cross-validation. These personalized models, which revealed significant prediction of mind wandering within each subject, were each explained by idiosyncratic whole-brain features. The default mode network, the network most widely associated with mind wandering, could be used to significantly predict mind wandering in all subjects, but functional connections important for prediction were variable across subjects. We then tested model generalizability across subjects using both cross-subject training of the personalized models generated by our analyses as well as two published group-level CPMs (stimulus-independent and task unrelated thought and sustained attention CPM). Neither the personalized models, nor the group-level models, significantly predicted mind wandering in all three subjects. These results suggest that the neural basis of self-reported mind wandering is variable across individuals and more broadly demonstrate the importance of applying individualized models in neuroimaging studies and clinical contexts.

Topic Area: ATTENTION: Other