Dynamic Functional Connectivity-Based Classification of Partial Sleep Deprivation Across Age Groups using Resting-State fMRI

Abstract

Sleep deprivation impacts large-scale brain network dynamics, and there are only limited neuroimaging biomarkers available to reliably detect sleep loss across individuals and age groups. In this study, Dynamic functional connectivity (DFC) analysis of resting-state fMRI was employed to classify partial sleep deprivation (PSD) versus full sleep across age groups. Using the Sleepy Brain Project I dataset (n=76 after data preprocessing), time-resolved functional connectivity features were extracted from 48 cortical regions via sliding-window analysis (window=99 TRs, stride=8) optimized through information-theoretic criteria. A hybrid feature selection combining stability selection and mutual information reduced 3,389 features to 50 discriminative markers. Linear SVM achieved ROC-AUC=0.85 (71% accuracy) across all participants. Remarkably, older adults showed 80% accuracy, exceeding the previous benchmark of 68%. Cross-age generalization maintained ROC-AUC=0.78-0.79. Permutation importance revealed distributed network alterations rather than localized changes. However, model scores showed no correlation with subjective sleepiness (ESS: r=-0.106, p=0.36), suggesting objective but not experiential capture of sleep loss. This study demonstrates that cross-subject classification of partial sleep deprivation vs complete sleep is achievable using rs-fMRI utilizing dynamic connectivity features, with accuracy levels superior to previous studies.