Ioannis Krillis, Theodore Antonakopoulos and George Kostopoulos:

An Automated Sleep Spindles Detection Tool and its Use on Spindles Analysis and Parameterization

The 28th Meeting of the Hellenic Society for Neuroscience - HSN2019, Crete, Greece, October 4-6, 2019.

Abstract: Spindles along with k-complexes are bursts of oscillatory brain activity on the thalamocortical system that occur during sleep. The accurate detection and analysis of these EEG rhythms is important in sleep studies. During a typical sleep hundreds of spindles occur in various EEG electrodes and the manual detection of all these spindles by an expert is a laborious and tedious task, so the need for a reliable software tool is profound. In this work we present such an automated software tool for the reliable detection of spindles and its use in the framework of a general software environment for spindles analysis and parameterization. The tool uses publicly available and custom spindle detection methods and analyses sequentially all channels of an EEG in order to identify the start and end of each spindle. Then the part of the signal that has been identified as a spindle is analyzed and a number of parameters is extracted so that each spindle is represented as a set of parameters and finally the whole sleep is represented as a dataset that can be further analyzed using automated machine learning techniques. Reliable spindle detection is achieved by using the results of multiple detection methods. The parameters of each method (i.e. detection threshold) are optimized through an iterative process for any given EEG channel and then each method is applied using its optimized parameters. The results of all applied methods are combined for achieving the final spindle detection. Experimental results based on artificial EEG signals demonstrate the validity of the proposed approach and results from the analysis of real EEG signals will be presented .