OpenSpindleNet: An Open-Source Deep Learning Network for Reliable Sleep Spindle Detection

Sleep spindles, an oscillatory brain activity occurring during light non-rapid eye movement (NREM) sleep, are important for memory consolidation and cognitive functions. Accurate detection is important for understanding the role of spindles in sleep state physiology and brain health and for better understanding sleep and neurological disorders. However, manual spindle labeling of electroencephalography (EEG) data is time-consuming and impractical for most clinical and research settings and intracranial EEG (iEEG) presents additional challenges for spindle identification due to its unique signal characteristics and recording environment.

This study introduces a novel, precise, and automatic spindle detection method for iEEG using a dual-head architecture to enhance performance, robustness, and ease of use. Our approach achieves a detection F1 score of 0.67 on a challenging iEEG dataset and 0.69 on the publicly available scalp EEG DREAMS dataset. Compared to existing methods such as SUMO, A7, and YASA, our model demonstrates superior performance in detecting, segmenting, and characterizing sleep spindles. This model contributes to open science and advances automated sleep spindle classification in iEEG. This will advance the development of more precise diagnostic and research tools and facilitate a deeper understanding of the role of sleep spindles in cognitive processes and neurological health.