Font Size: a A A

RBD Detection And Aided Diagnosis Based On Eeg Features And Depth Network

Posted on:2022-11-06Degree:MasterType:Thesis
Country:ChinaCandidate:S LiuFull Text:PDF
GTID:2504306611486054Subject:Telecom Technology
Abstract/Summary:PDF Full Text Request
Parkinson’s disease(PD)is the second most common neurodegenerative disease.Most PD patients will develop as rapid eye movement sleep behavior disorder(RBD).As a non-tremor motor subtype,PD patients with RBD have greater risk of various complications than simple PD patients.Correctly identifying RBD in PD has therapeutic and prognostic significance.At present,the existing diagnosis methods of PD with RBD have the problems of low equipment comfort and low accuracy.This paper proposes an intelligent aided diagnosis method of PD with RBD based on EEG multi-level fusion features and depth network,to improve the shortcomings of current diagnostic methods.Firstly,according to patient’s sleep monitoring reports and EEG data in the polysomnography,five minutes awake original EEG signals before sleeping of PD with RBD and PD without RBD subjects are filtered and segmented to retain the bands of interest and optimize the computational complexity.Each subject’s six-channels EEG signals are divided into 150 equal time window signals.Secondly,discrete wavelet transform is used to realize the multi-level timefrequency feature express of EEG signals,and the statistical,spectral and non-linear features are extracted to construct multi-level fusion feature space.Finally,the deep network classification model is constructed based on preprocessing,multi-level fusion feature extraction,and bidirectional long-short term memory network combined with three-fold cross validation,and the overfitting is improved by dropout layer.The final diagnosis of subjects is realized by postprocessing the classification results of the classification model.The classification accuracy of the proposed RBD detection method based on time window is 86.50% and the AUC value is 0.95.Compared with the traditional classification algorithms and existing literature method,the accuracy of the proposed detection method is improved by 7.42% and 14.79% respectively,and is equivalent to the AUC value of automatic analysis of polysomnography signal.In addition,the effectiveness of the proposed aided diagnosis method is verified based on the external test set,and the diagnosis accuracy based on subjects after post-processing is 87.50%,which realizes the aided diagnosis of PD with RBD.Although it is slightly lower than the accuracy of diagnosing RBD with polysomnography signal,it is better than the results of clinical questionnaire.The proposed method has high clinical application value.
Keywords/Search Tags:EEG features, Parkinson’s disease, deep network, rapid eye movement sleep behavior disorder
PDF Full Text Request
Related items