Abstract:Tracking how individual human brains change over extended timescales is crucial to clinical scenarios ranging from stroke recovery to healthy aging. The use of resting state (RS) activity for tracking is a promising possibility. However, it is unresolved how a person's RS activity over time can be decoded to distinguish neurophysiological changes from confounding cognitive variability. Here, we develop a method to screen RS activity changes for these confounding effects by formulating it as a problem of change classification. We demonstrate a novel solution to change classification by linking individual‐specific change to inter‐individual differences. Individual RS‐EEG was acquired over five consecutive days including task states devised to simulate the effects of inter‐day cognitive variation. As inter‐individual differences are shaped by neurophysiological differences, the inter‐individual differences in RS activity on one day were analyzed (using machine learning) to identify distinctive configurations in each individual's RS activity. Using this configuration as a decision‐rule, an individual could be re‐identified from 2‐second samples of the instantaneous oscillatory power spectrum acquired on a different day both from RS and confounded‐RS with a limited loss in accuracy. Importantly, the low loss in accuracy in cross‐day vs same‐day classification was achieved with classifiers that combined information from multiple frequency bands at channels across the scalp (with a concentration at characteristic fronto‐central and occipital zones). Taken together, these findings support the technical feasibility of screening RS activity for confounding effects and the suitability of longitudinal RS for robust individualized inferences about neurophysiological change in health and disease.

Robustness of individualized inferences from longitudinal resting state EEG dynamics