Sleep Spindles as Biomarkers for Idiopathic REM Behaviour Disorder

Sleep spindles are key markers of non-rapid eye movement sleep, with significant applications in understanding the pathophysiology of sleep-related and neuropsychiatric disorders. However, annotating spindles in sleep EEG recordings is a labor-intensive and time-consuming task. To address this issue, this study proposes a machine-assisted framework for spindle detection and explores the diagnostic utility of spindle dynamics as biomarkers for idiopathic REM behavior disorder (iRBD). A total of 18 hours of EEG recordings from 17 patients with iRBD conditions were annotated to develop the spindle detection system. Classification models were trained using both handcrafted features and raw EEG signals. A Long Short-Term Memory (LSTM) network model achieved the highest unweighted average recall (UAR) of 83.31% for spindle detection on test data. The automatic spindle detection system was subsequently applied to a separate dataset of 214 subjects (181 healthy controls and 33 iRBD cases) to extract spindle dynamics for iRBD identification. While the model achieved high specificity (0.82) and precision (0.89) for healthy controls, sensitivity for iRBD detection was moderate (0.45), resulting in an overall accuracy of 0.77.Clinical relevance- The proposed system could aid early diagnosis of iRBD by clinicians, enabling timely interventions and improved management strategies for at-risk individuals.