Comprehensive Summary
OpenSpindleNet, an open-source software developed by Sejak et al., uses deep learning methods to detect sleep spindles from scalp EEG and intracranial EEG recordings. Sleep spindles are short bursts of neural activity in the range of 11-16 Hz that occur during stage N2 of non-REM sleep. Abnormalities in these patterns can have clinical significance, as these spindles are thought to be associated with learning and memory. Sejak et al. introduce a new generalized open deep-learning model to better detect sleep spindles, particularly in intracranial EEG (iEEG). The main differences between Sejak et al. 's model and other spindle detection models are this model’s focus on improving iEEG detection and its dual-head technique that separates the identification and segmentation tasks of detection. In this study, Sejak et al. compared their model’s predictions to those of other common models, and their model performed better on all metrics for the iEEG data, and for almost all metrics for the scalp EEG data.
Outcomes and Implications
Traditionally, sleep spindles are identified manually, which is time consuming and introduces inter-recorder unreliability. From a clinical perspective, spindle activity abnormalities can be associated with various neurological and psychiatric disorders, such as epilepsy, schizophrenia, autism, and neurodegenerative diseases, and may be able to be used as biomarkers for these in future work. Changes in sleep spindles may also give insight into the neuronal networks involved in learning and memory. Improving spindle detection in iEEG may prove essential in that network research, as iEEG records from deeper cortical regions than scalp EEG and gives a more localized view of the neural activity. OpenSpindleNet provides a more accurate model than current models for spindle detection, particularly in iEEG.