Validating an automated sleep spindle detection algorithm using an individualized approach

Abstract

SUMMARYThe goal of the current investigation was to develop a systematic method to validate the accuracy of an automated method of sleep spindle detection that takes into consideration individual differences in spindle amplitude. The benchmarking approach used here could be employed more generally to validate automated spindle scoring from other detection algorithms. In a sample of Stage 2 sleep from 10 healthy young subjects, spindles were identified both manually and automatically. The minimum amplitude threshold used by the prana Ò (PhiTools, Strasbourg, France) software spindle detection algorithm to identify a spindle was subject-specific and determined based upon each subjectÕs mean peak spindle amplitude. Overall sensitivity and specificity values were 98.96 and 88.49%, respectively, when compared to manual scoring. Selecting individual amplitude thresholds for spindle detection based on systematic benchmarking data may validate automated spindle detection methods and improve reproducibility of experimental results. Given that interindividual differences are accounted for, we feel that automatic spindle detection provides an accurate and efficient alternative approach for detecting sleep spindles. k e y w o r d s automated detection, individual differences, sleep, sleep spindles

Quantifying chirp in sleep spindles

“…This is justified by the fact that there is still debate over an optimal voltage threshold for automatic spindle detection (Huupponen et al, 2000;Bodisz et al, 2009;Ray et al, 2010). Therefore, the NREM background atom dataset is assumed to represent a mixture of activities in the sigma frequency range, including elements that might correspond to false positive spindle detections when one considers the visual criterion as the gold standard.…”

An adaptive neuro-fuzzy inference system for sleep spindle detection

“…In recent years, several spindle detection methods were proposed [6][7][8]11], decision tree with three features [8], Multi-Layer Perceptron (MLP) without feature extraction [6], and support vector machine (SVM) using 15 adaptive autoregressive (AAR) parameters as features [12]. The sensitivity among these studies ranges from 70% to 98.96%, the specificity ranges from 88.49% to 98.76%, while the false positive rate ranges from 2.85% to 37.2%.…”

Analysis of EEG Sleep Spindle Parameters from Apnea Patients Using Massive Computing and Decision Tree

“…However, this also increases computational cost. When using a clinical EEG sample, subject inter-variability becomes clear and makes necessary a selection approach based on a spindle amplitude criterion adjustable for individual and channel [13].…”