Classification Of Depression EEG Signals Based On Topological Data Analysis

Depression is a global mental and emotional disorder that seriously affects the health and quality of life of patients,and also brings a huge economic burden to society.Electroencephalogram(EEG)signals contain rich physiological and pathological information,so more and more researchers are applying it to the field of intelligent recognition of depression.Topological Data Analysis(TDA),which combines topology theory with statistical analysis methods,is a new data analysis tool and has been used to extract EEG features.However,apart from the commonly used continuous homology entropy features,there are relatively few other available topological features.Based on this,this article has done the following work:Firstly,this article proposes a novel topology persistence feature based on persistence barcode diagram.First of all,the input of topological data analysis is realized through the brain function connection matrix,and then the Vietoris Rips simplicial complex is continuously constructed.The continuous bar code graph is obtained through the continuous homology analysis,and the birth and death time data of the continuous homology information are extracted.In order to more fully express the continuous information in barcode images,this article considers combining statistical indicators in statistics and further proposing Carlsson II topological features based on Carlsson topological features.The Carlsson II feature and the common continuous homology entropy feature are calculated on the MODMA and EDRA datasets,and they are input into the support vector machine,random forest and K-nearest neighbor classifiers.Finally,the classification experimental results are compared.The results show that the Carlsson II features proposed in this article have achieved effective improvements compared to the Carlsson topological features,and the classification results obtained through Carlsson II features are better than those obtained through continuous homology entropy features.Secondly,this article constructs a CNN+PLLay model for intelligent recognition of depressive EEG signals.Introducing a Topological Layer based on Persistence Landscapes(PLLay)in the Convolutional Neural Network(CNN)model,enhances the learning ability of the CNN model by providing important topological features of data structures in the network.The PLLay topology layer in the model is the key to feature extraction,and in this layer,a continuous scatter plot is formed through continuous homology analysis.Due to the low computational efficiency of continuous scatter plots,it is difficult to directly apply them to classifier learning.Therefore,this article uses a segmented function to transform it into a more suitable continuous landscape graph for metric learning,and reflects the topological structure characteristics of the data by weighting the continuous landscape.The effectiveness of the constructed CNN+PLLay model was verified on the MODMA and EDRA datasets.The new model achieved classification accuracy of 97.88% and 97.65% respectively,significantly improving the accuracy of using topological data information for depression EEG signal classification.