基于大众外包方法的纺锤波检测可行性分析

发布时间：2018-01-18 10:18

本文关键词：基于大众外包方法的纺锤波检测可行性分析　出处：《西安电子科技大学》2015年硕士论文　论文类型：学位论文

【摘要】：睡眠纺锤波(Spindle)是非快速眼动睡眠(NREM)N2阶段的标志,它的频率范围为11~16Hz,产生时间大于0.5秒,振幅先增大后减小,形状类似于梭子。纺锤波在人的记忆和智力的机制研究中,以及在一些精神类疾病的临床诊断上都有重要作用。到目前为止,纺锤波自动检测算法层出不穷,而基于肉眼观测的手动划分一直以来都是准确率最高的方法,同时专家的肉眼手动划分被称为纺锤波检测的金标准,但是专家一般需要经过专业培训而且很难找到,所以本文通过大众外包的方法获得了大量非专家标记的纺锤波数据集,然后得到非专家组的标准,并将其与专家组的金标准进行对比,来看看是否可以通过非专家组的标准来代替专家组的金标准。在进行本文的实验之前,我们需要采集并选取一些实验所用的睡眠脑电数据。在采集脑电数据之前,需要被试填写一些量表来对被试近期的情绪、睡眠以及其他情况进行简要评估,所以开发了在线睡眠研究问卷调查系统。该系统代替了传统的纸质问卷调查,一方面使得作用范围更广、速度更快,另一方面可以节省很大的人力、财力、物力和时间,而且可以将调查的数据全部存储于计算机中,以便随时使用。由于一个人一晚上的睡眠脑电数据量很大,而且本研究是通过大众外包的方法来标记纺锤波的,所以参与标记纺锤波的被试会很多,因此最终需要处理的数据量会很大。如果采用MATLAB离线方式来标记纺锤波的话,后期数据统计和数据处理都会比较繁琐,所以开发了基于WEB的在线检测纺锤波系统。那么只需要被试身边有一台电脑就可以在线标记纺锤波,然后将标记的结果存储到远程数据库。对专家组和非专家组通过在线检测纺锤波系统标记的数据集进行分析,可以得到以下结论:对于专家组来说,组阈值T-group为0.3和重叠阈值T-overlap为0.45时,此时专家组金标准是最优的。所有的专家与专家组金标准比较,其平均表现为0.84007±0.023(均值±方差),这表明每位专家与专家组金标准具有较好的一致性。对于非专家组来说,组阈值T-group为0.35和重叠阈值T-overlap为0.3时,非专家组标准是最优的。所有的非专家与非专家组标准进行比较,其平均表现为0.7246±0.1008(均值±方差),显然与专家组的平均表现相比,非专家组平均表现的均值变小,方差变大。这说明非专家之间的一致性不如专家的高。非专家组标准与专家组金标准比较的F1值为0.7557,也就是说虽然非专家之间的一致性没有专家之间的高,但是非专家组标准与专家组金标准的差别程度还是可以接受的,即由非专家组标准代替专家组标准可行的。同时还将非专家组标准和RMS自动检测算法进行对比,发现非专家组标准是优于RMS自动算法的。
[Abstract]:Sleep spindle is a non-REM sleep NREMN _ 2 stage marker, its frequency range is 114Hzand the time of generation is more than 0.5 seconds. The amplitude increases first and then decreases, similar to the shape of the shuttle. Spindles play an important role in the study of the mechanism of human memory and intelligence, as well as in the clinical diagnosis of some mental disorders. So far. Automatic spindle wave detection algorithms emerge one after another, and manual partition based on naked eye observation has always been the most accurate method, and the expert manual partition is called the gold standard of spindle wave detection. However, experts usually need professional training and are difficult to find, so this paper obtains a large number of non-expert mark spindle wave data set through the method of public outsourcing, and then get the standard of non-expert group. And compare it with the gold standard of the expert group to see whether the gold standard of the expert group can be replaced by the standard of non-expert group. We need to collect and select sleep EEG data used in some experiments. Before we collect EEG data, we need to fill out a number of scales to briefly assess the participants' recent mood, sleep and other conditions. Therefore, an online sleep research questionnaire system has been developed. This system replaces the traditional paper questionnaire. On the one hand, it makes the function wider and faster, on the other hand, it can save a lot of manpower and financial resources. Material resources and time, and all the data can be stored in the computer, in order to use at any time. Because a person's sleep EEG data volume is very large. And this study is through the mass outsourcing method to mark the spindle wave, so many participants involved in marking spindle wave. Therefore, the amount of data to be processed will be very large. If the MATLAB off-line method is used to mark the spindle wave, the later data statistics and data processing will be more cumbersome. Therefore, an on-line spindle wave detection system based on WEB is developed, and only a computer is needed to mark the spindle wave online. The results of marking are then stored in the remote database. By analyzing the data sets of the expert group and the non-expert group through the on-line detection of the marking of the spindle wave system, the following conclusions can be drawn: for the expert group. When the group threshold T-group is 0.3 and the overlap threshold T-overlap is 0.45, the expert group gold standard is optimal. All experts are compared with the expert group gold standard. Its average performance is 0.84007 卤0.023 (mean 卤variance), which indicates that each expert has good consistency with expert group gold standard. When the group threshold T-group was 0.35 and the overlap threshold T-overlap was 0.3, the non-expert group criterion was the best. All the non-experts were compared with the non-expert group standard. Its average performance is 0.7246 卤0.1008 (mean 卤variance), obviously compared with the average performance of the expert group, the average performance of the non-expert group is smaller. The variance increases. This shows that the consistency among non-experts is not as high as that of experts. The F1 value of the non-expert group standard compared with the expert group gold standard is 0.7557. That is, although the consistency among non-experts is not as high as that among experts, the difference between the non-expert group criteria and the expert group gold criteria is acceptable. That is to say, it is feasible to replace the expert group standard with the non-expert group standard. At the same time, the comparison between the non-expert group standard and the RMS automatic detection algorithm shows that the non-expert group standard is superior to the RMS automatic algorithm.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：R740

【相似文献】