Decoding ruminative states from neurophysiological patterns.

Individuals with depression often engage in iterative "rumination" about challenging situations and potential outcomes. Although the state of rumination has been associated with diverse univariate neurophysiological features, the potential to use multivariate patterns to decode it remains uncertain. In this proof-of-principle study, we trained participant-specific linear support vector machines to differentiate state rumination from distraction using patterns in the alpha, beta, and theta bands, as well as inter-channel connectivity. We used validated tasks to induce rumination or distraction for eight minutes in 24 depressed individuals in six runs over three sessions. During inductions, we recorded 64-channel EEG data and measured self-reported levels of rumination. Participants reported strongly increased rumination, and averaged across all participant-specific decoders we classified state rumination from EEG patterns with small albeit significant accuracy. However, on an individual level, decoding reached accuracies above chance level in only 10 out of 24 participants with heterogenous feature weights across individuals. Our study demonstrates that ruminative states can be decoded from inter-individually heterogenous neurophysiological patterns, but variability in ruminative processes limits the generalization of decoding approaches across participants.