Optimization Of Sleep Spindle Detection Algorithm And Comparative Analysis Of Features

In recent society,about 25%people are undergoing from sleep disorders in the world,but most of the diagnosis of sleep disorders still rely on the sleep quality scale analysis,lacking of an objective basis for disease diagnosis.With the development of electroencephalogram(EEG),the spindle,a form of transient oscillating brainwave,was discovered and described by sleep researchers.The researchers have found effective distinctions in spindles among people of different age,different gender and different brains.However,the domestic research on the relationship between sleep disorder and spindles is not deep enough at present.Sleep spindles detection is one of the focuses of spindles research.The traditional method of detection is manual detection by experienced clinician.This method is a waste of time and labor,so researchers have proposed lots of spindle automatic detection algorithms.However,many algorithms do not perform well in some evaluation indexes,and show some capability distinctions in different data sets.For purpose of resolve these issues,we promote the algorithm of automatic detection of sleep spindles and break down the spindles traits of somnipathy and normal people.Through the research of many kinds of conventional spindles automatic exploration algorithms,this paper innovatively presents a multi-algorithm amalgamation spindles automatic exploration algorithm to promote the capability of spindles automatic exploration.The amalgamation algorithm uses Morlet wavelet Algorithm,RMS algorithm and optimized k-means cluster.Based on the sleep disorder data set we collected,the average precision of the raised fusion algorithm is91.6%.The average performance of single detection algorithm Morlet wavelet on precision and recall is 86.7%and 55.2%respectively,while that of RMS algorithm is75.9%and 85.0%respectively.Contrasted with the classical algorithm,the capacity of the raised fusion algorithm is obviously changed.It been proved that the fusion algorithm is still stable in the normal population data set.After the automatic spindle detection was completed,we performed a Pearson correlation analysis of Pittsburgh sleep quality index scores and spindle characteristics in each sample.Between sleep quality scores and spindle density,thep=1.84*10^-8.This suggests that the higher the density of spindles during sleep,the better the quality of sleep.Similarly,between sleep quality scores and spindle amplitude,the p=1.33*10^-4.The mean amplitude in general people is slightly higher than that in people with sleep disorders.However,the difference in frequency characteristics is not particularly obvious,the p=0.667 between frequency and sleep scores.There was no conspicuous connection.In addition,there was no conspicuous discrepancy in the amount of spindles among the two groups in symmetric channels C3/C4 and F3/F4.The laboratory results show that the capacity of the raised algorithm is content.At the same time,the difference of the density and amplitude of spindles can be a reference feature for the diacrisis of sleep diseases and supply some objective evidence for clinic.