好色先生

好色先生

Explore the latest content from across our publications
Join The AAN

Log In

Forgot Password?
Create New Account

Loading... please wait

Abstract Details

An EEG-Based Machine Learning Classifier To Differentiate Hippocampal-Coupled Sleep Spindles In Humans
Epilepsy/Clinical Neurophysiology (EEG)
P16 - Poster Session 16 (8:00 AM-9:00 AM)
10-001

Develop a classifier to identify sleep spindles that are coupled with hippocampal ripples using only EEG data as a biomarker for sleep-dependent memory consolidation.
The triple coupling of thalamocortical sleep spindles, cortical slow oscillations (SOs), and hippocampal sharp-wave ripples is implicated in sleep-dependent memory consolidation. Measuring this coupling as a biomarker for memory consolidation would be a useful for case control studies and clinical trials but hippocampal activity cannot be reliably detected from the scalp. Here we developed a machine learning classifier to identify spindles that are coupled with ripples using EEG data alone.
Four inpatients with epilepsy with simultaneous scalp and intracranial EEG with hippocampal contacts, were recruited from the MGH EMU. SOs, spindles, and ripples were detected during NREM sleep using validated algorithms. Coupling was defined as the temporal co-occurrence of the oscillations. A random forest classifier was developed to categorize spindles into four classes (non-coupled, SO-coupled, ripple-coupled, or triple-coupled) using scalp EEG spindle and SO characteristics (including peak amplitude, duration, and frequency). True classes based on intracranial recordings were used to train the classifier and held-out data were used to test the classifier.
True ripple-coupled spindles occurred at a rate of 2.3/min (range: 1.2-3.7) and triple-coupled spindles occurred at a rate of 0.26/min (range: 0.13-0.56).  The classifier has a geometric mean, which combines sensitivity and specificity, of .70 [range 0:1, with 1 being perfect classification] in differentiating between the four spindle classes.
A machine learning classifier can successfully identify ripple-coupled spindles using scalp EEG alone. We expect its performance to improve as it is trained on a larger dataset. This work will allow us to determine whether coupled spindles are a superior biomarker of sleep-dependent memory consolidation than all spindles (the standard measure) and may provide a better target for treatment development for memory deficits.
Authors/Disclosures
Bryan Baxter, PhD (Massachusetts General Hospital)
PRESENTER 		Dr. Baxter has received research support from 好色先生/American Brain Foundation/The McKnight Brain Research Foundation.
No disclosure on file
No disclosure on file
No disclosure on file
No disclosure on file
No disclosure on file
No disclosure on file
Catherine J. Chu, MD (Massachusetts General Hospital / Harvard Medical School) Dr. Chu has received personal compensation in the range of $5,000-$9,999 for serving as a Consultant for Biogen Inc. The institution of Dr. Chu has received research support from Biogen Inc. The institution of Dr. Chu has received research support from NIH.
Dara S. Manoach, MD (Beth Israel Hospital) The institution of Dr. Manoach has received research support from NIMH. The institution of Dr. Manoach has received research support from MGH ECOR.