What makes our brain unique: Insights from a two-night fMRI-EEG sleep study.
Fan Nils Yang, Dante Picchioni, Jacco A De Zwart, Peter Van Gelderen, Jeff H. Duyn
Imaging neuroscience (Cambridge, Mass.) 2026 DOI: 10.1162/imag.a.1333 (opens in new tab)
Study at a glance
AI-extracted from the abstract| Characteristics | Observational cohort Peer reviewed |
|---|---|
| Population | Healthy participants |
| Keywords | N3 stage Brain fingerprints Deep learning Dynamic functional connectivity Hippocampus |
| Key findings | FC fingerprinting remained robust and improved during N3 sleep relative to wakefulness, with hippocampal connectivity playing a larger role during N3. A deep-learning classifier achieved high identification accuracy, and models trained on deep-sleep data generalized well across all stages, favoring the view that intrinsic network architecture underlies FC fingerprints. |
Abstract
Functional connectomes (FCs) derived from fMRI are unique to an individual and can be reliably distinguished from those of others. However, the neurobiological basis for this "fingerprint"-like uniqueness remains unclear. One hypothesis holds that FC fingerprinting reflects idiosyncratic conscious experiences, while an alternative suggests that stable, intrinsic brain functional network structure underpins individual specificity. To test these hypotheses, we analyzed FC fingerprints across sleep stages using a two-night concurrent EEG-fMRI dataset from healthy participants and exploited the natural fluctuations in consciousness that occur across the sleep-wake cycle: during sleep, consciousness gradually diminishes from N1 to N3 non-REM (NREM) sleep, then partially reemerges during REM sleep. Contrary to the consciousness-dependent hypothesis, we found that FC fingerprinting remained robust, and in fact improved during deep, unconscious sleep (N3) relative to wakefulness. Moreover, the brain networks contributing most to individual identification shifted across sleep stages, with hippocampal connectivity playing a larger role during N3. Finally, using a deep-learning classifier (multi-layer perceptron) on short (90 seconds) fMRI segments, we demonstrated high identification accuracy; notably, models trained on deep-sleep data generalized well across all stages, including wakefulness. These findings favor the view that individual-specific network architecture rather than moment-to-moment conscious experience underlies FC fingerprints, with important implications for their use as biomarkers in both healthy and clinical populations.