Real-time detection of N2 and N3 sleep from scalp and intracranial EEG

Researchers have developed a system to detect N2 and N3 sleep stages in real-time using scalp and intracranial EEG. These sleep stages are crucial for memory enhancement through closed-loop stimulation. Conventional methods label 30-second epochs of EEG, which are not suitable for real-time applications. The new system estimates the current sleep stage every 2 seconds from the previous 10 seconds of EEG data.

In a study involving six adults with drug-resistant epilepsy, the real-time detection matched offline scoring of the same recordings, with an area under the curve (AUC) of 0.953 for scalp EEG and 0.958 for a simulated rule. Thalamic channels alone achieved an AUC of 0.934, surpassing mesial temporal channels in every patient. The scalp tracker, when retrained on 28 healthy adults, reached an AUC of 0.976.

When patients entered N2+N3 sleep, the system allowed stimulation within 5 minutes in 63-73% of transitions, with a median delay of 42-64 seconds. After exiting N2+N3 sleep, stimulation was disabled after a median of 19-22 seconds. However, the system incorrectly switched on 1.4-1.8 times per hour outside N2+N3 sleep and remained active in 28-31% of REM sleep, compared to just 2% in healthy adults.

These findings demonstrate the potential of both scalp and thalamic EEG for real-time N2+N3 detection, potentially enabling sleep-stage-dependent stimulation in implanted neuromodulation devices. Scalp EEG could also be utilized for noninvasive neuromodulation. The authors suggest evaluating closed-loop systems based on both sleep-stage classification accuracy and their ability to reliably and promptly enable and disable stimulation.