Machine-Learning-Based Classification of Acute Partial Sleep Deprivation with Resting-State fMRI

Loading...
Thumbnail Image

Date

2024-12-06

Authors

Yang, Xi

Journal Title

Journal ISSN

Volume Title

Publisher

University of Oregon

Abstract

Insufficient sleep is highly prevalent. Limited knowledge has been accrued on the functional correlates of acute partial sleep deprivation in the awake brain. As resting-state functional magnetic resonance imaging (rs-fMRI) becomes an essential measure to investigate spontaneous neural activity and intrinsic functional connectivity, applying machine learning to rs-fMRI to classify the state of acute partial sleep deprivation remains an uncharted area. In the present study, based on sleep deprivation literature, a set of predetermined rs-fMRI region and network functional connectivity features were used to classify the sleep states (sleep deprived/well-rested) of the senior (N = 34, age 65-75) and young adult (N = 41, age 20-30) participants in an archival dataset. The best performing support vector machine model classified the sleep states of the senior adult participants with a 68% accuracy rate. During external validation, this model trained on senior adults demonstrated low transferability to the young adult dataset. Low classification accuracy were reported in models trained on young adult dataset. The theoretical implications of the findings and recommendations for future research were discussed to contribute to a multi-modal understanding of the mechanism of sleep insufficiency as a causal factor of neural vulnerability and inform neurobehavioral interventions.

Description

Keywords

Citation