睡眠是人类至关重要的生命活动，睡眠问题可能会引发各种身体和精神类疾病。睡眠障碍是帕金森病（Parkinson’s disease, PD）患者常见的非运动症状之一，严重影响了患者的生活质量，因此PD患者的睡眠研究具有重要意义。睡眠分期是睡眠研究的基础，目前关于PD患者的自动睡眠分期研究较少。脑深部电刺激（Deep Brain Stimulation，DBS）技术是目前临床上治疗难治性PD的一种有效方法，可以有效改善患者的运动症状，但DBS技术能否从客观上改善睡眠目前仍没有定论。基于这一背景，本文对已完成DBS植入的PD患者的睡眠监测数据展开了探究。首先，对PD患者进行了睡眠脑电特征分析。使用多窗谱法进行功率谱估计，提取了一组有效的分类特征集，并对文献中的方法进行了部分改进，实现了纺锤波、K复合波和慢波的提取。在此基础上，对不同通道、PD患者与健康人和DBS植入不同时间后睡眠脑电的频谱和纺锤波特征进行了对比分析，发现了PD患者存在 波段的额叶优势；与健康人相比，PD患者清醒期 和 波功率更低；但随着DBS植入时间增加，清醒期的 波功率和睡眠期的 波功率会增加，且大部分患者的纺锤波峰峰值升高。然后，基于提取的分类特征集进行了自动分期方法研究，设计了一个支持向量机（Support Vector Machine , SVM）多分类器，并加入了启发式规则进行分类结果的修正。测试结果表明，当睡眠分期分辨率为5s/epoch时，分类器总平均准确率达到82.5%。最后，设计了一个基于睡眠自动分类器的图形用户界面（Graphical user interface，GUI），对前面研究工作进行了集成，能够实现数据可视化、数据预处理、脑电分析和睡眠自动分期等功能。综上，本文一方面探究了PD和DBS刺激对睡眠脑电的影响，另一方面基本实现了基于PD患者PSG数据的睡眠自动分期，但分类器的泛化能力还有待进一步提高，未来可以从睡眠的连续性信息方面寻找更优的启发式规则，对于不同样本巨大的差异性，如何从根本上提高整体的泛化能力或许将是未来睡眠自动分类器能否广泛应用的关键。
Sleep is one of the most important life activities of human beings, and sleep disorders can cause various physical and mental health problems. Sleep disorder is one of the common non-motor symptoms of Parkinson’s disease (PD) patients, which seriously affects the quality of life of patients. Therefore, sleep research in PD patients is of great significance. Sleep staging is the basis of sleep research. At present, there are few studies on automatic sleep staging for PD patients. Deep Brain Stimulation (DBS) technology is currently an effective clinical treatment for refractory PD, which can effectively improve the patient's motor symptoms, but whether DBS technology can objectively improve sleep is still uncertain.Based on this background, this article explores the sleep monitoring data of PD patients who have completed DBS implantation. First of all, it focused on the analysis of the sleep EEG characteristics of PD patients, and used the multi-window spectrum method for spectrum estimation to obtain effective classification features. And by partially improving the method in the literature, the spindle wave, K-complex wave and slow wave are extracted successfully. And further compared the frequency spectrum and spindle wave characteristics of different channels, PD patients and healthy people and DBS implanted for different time, it is found that PD patients have the advantage of the frontal lobe of the delta band. In addition, compared with healthy people, PD patients have lower beta and delta wave power during the awake phase. However, as the DBS implantation time increases, the beta wave power in wake phase and the delta wave power during sleep becomes larger and larger, and the peak-to-peak value of the spindle in most patients will increase.Then, based on the feature set extracted from the study of sleep EEG characteristics, the sleep staging method was studied, and a Support Vector Machine (SVM) multi-classifier was designed, and heuristic rules were added to modify the classification results. The test results show that when the resolution of sleep staging is 5s/epoch, the total average accuracy of the classifier reaches 82.5%.Finally, a graphical user interface based on the sleep automatic classifier is designed, which integrates the previous research work and can realize the functions of data visualization, data preprocessing, EEG analysis and automatic sleep staging.In summary, on the one hand, this article explores the impact of PD and DBS stimulation on sleep EEG, on the other hand, it basically realizes the automatic sleep staging based on PSG data of PD patients, but the generalization ability of the classifier needs to be further improved. Looking for better heuristic rules from the continuity of sleep information, for the huge differences of different samples, how to fundamentally improve the overall generalization ability may be the key to whether automatic sleep classifiers can be widely used in the future.
