Abstract

AbstractBackgroundDementia is a growing cause of disability and loss of independence in the elderly, yet remains largely under‐diagnosed. Early detection and classification of dementia may help close this diagnostic gap and improve management of disease progression. EEG sleep patterns have been identified as a potential biomarker to detect Alzheimer’s disease and other neurodegenerative diseases.MethodFrom a dataset of 9834 polysomnograms, sleep architecture and microstructure features such as frequency band powers, EEG coherence, and spindle density were extracted. Patients were labeled as belonging to dementia, mild cognitive impairment (MCI), or cognitively normal (CN) groups based on clinical diagnosis, Montreal Cognitive Assessment (MoCA), Mini‐Mental State Exam (MMSE) scores, Clinical Dementia Rating (CDR) and medications. We trained logistic regression, random forest, and XGBoost models to classify patients into Dementia, MCI, and CN groups.ResultNested cross validation results show an AUC of 0.81 (F1 = 0.76) for binary classification of dementia vs CN groups and a mean AUC of 0.75 (F1 =0.57) for multiclass classification of dementia vs. MCI vs CN groups. REM latency, spindle activity, duration, frequency, slow wave oscillation, delta/alpha band power in wake, N1 theta/alpha band power in N1 were among the top weighted features.ConclusionOur dementia classification algorithms show promise for incorporating dementia screening techniques into routine sleep EEG and providing diagnostic, monitoring, and prognostication capabilities.

Full Text
Paper version not known

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call