Background
Rumination, an aberrant form of spontaneous thoughts, is a common symptom of major depressive disorder and anxiety disorders. Developing non-intrusive tools to study the content and dynamics of spontaneous thoughts is likely to be helpful for understanding rumination. We developed a novel approach to decode spontaneous thoughts using functional magnetic resonance imaging (fMRI) data during resting state.
Methods
fMRI data of 20 healthy participants (8 females, mean age 31.2, SD=10.1) were acquired during two tasks: free spontaneous thoughts (resting state for 3-5 minutes) followed by prompted verbal report of immediate thoughts, watching the movie Forrest Gump. Shared response modeling (functional alignment) was used to map whole-brain activity of all participants to a low dimensional space in which neural signals are synchronized during movie watching. Topic modeling based on a large language model identified 9 major topics among reported thoughts. Neural activity patterns in the low-dimensional space corresponding to the same topic in the 10 seconds before prompts were averaged across participants. Cross-validated decoding accuracy of thought topics was evaluated based on the cosine similarities between neural patterns of left-out participants and the average patterns of each topic from other participants.