EEG Spike Detection In Children With BECT Epilepsy

The benign childhood epilepsy with centrotemporal spikes(BECT)is one of the most common epilepsy syndromes in childhood.The most obvious feature of BECT is the presence of indeterminate spikes in the Rolandic region of the brain during interictal periods.The frequency of the discharge is closely related to the patient’s condition.Therefore,the observation and data statistics of the spike discharge phenomenon have become an important basis for neurologists to BECT patients.Designing a spike detection algorithm for children’s EEG that can quickly and accurately locate the discharge phenomenon can assist doctors in the rapid diagnosis and treatment of BECT patients.The current mainstream spike detection algorithms do not further use the spike discharge characteristics of BECT syndrome to optimize the detection performance because their detection objects are not BECT children,and often ignore the specificity of children’s EEG,such as more interference and high amplitude,which makes these algorithms tend to have poor detection performance on BECT patient data.Aiming at the above problems,this thesis has carried out research,and the main work is divided into the following parts:1.A single-channel spike detection method based on EEG time series features and longshort-term memory neural network is proposed,which makes full use of the time series features of EEG during spike discharge to achieve high-performance spike detection on singlechannel EEG data.The method focuses on the characteristics of EEG as a kind of sequence data,and extracts the smooth nonlinear energy(SNE)and morphological filter characteristics(MC)from the EEG to describe the time domain sequence feature of the EEG data during the spike discharge period.And the recurrent neural network architecture,which has excellent performance in sequence data processing,is used for feature analysis and spike classification and detection.At the same time,the data interference and the unbalanced phenomenon of spike and non-spike data are also dealt with in the detection process.The method finally achieved mean F1 scores,sensitivity and precision of 88.54%,92.04% and 85.75% on the Children’s Hospital of Zhejiang University(CHZU)neurology dataset.All detection performance indicators are superior to the existing mainstream spike detection algorithms.2.A spike detection algorithm based on weighted fusion of EEG multi-channel data is proposed,and the multi-channel distribution information of spike discharge is used to further improve the detection performance and generalization ability of the method as a whole.The single-channel spike detection algorithm has low design complexity and flexible implementation,but multi-channel spike detection algorithms have natural advantages in terms of the amount of available information.The method firstly completes candidate sample screening and artifact removal through the waveform characteristics of spikes and the special phenomenon of ”peak-to-peak” in multi-channel EEG data.Then,according to the three most significant characteristics of spike discharge,single-channel data is generated by weighting based on the amplitude,waveform and source distance weights of each channel data of the multi-channel spike candidate samples.Combined with the proposed time series feature extraction and LSTM neural network classification method,a spike detection method with excellent detection performance and generalization performance is completed.On the CHZU dataset,compared with the proposed single-channel spike detection algorithm,the F1 score,sensitivity and accuracy of the method on the single-person dataset have been further improved by 7.43%,4.15%,and 10.08%.On the mixed dataset,the detection performance of this method does not decrease significantly,which shows the excellent model generalization ability of this method,and all detection performance indicators have obvious advantages over the existing mainstream spike detection algorithms.3.Developed a children spike detection system on multi-channel EEG.The system was developed on the MATLAB GUI platform,including six modules: data import,data processing,sample screening,artifact removal,data fusion and spike detection.Based on the pretrained neural network classifier and hyperparameters,the program can complete the spike detection process from data import to spike detection.It also includes multiple result display windows and a series of interactive designs,which can help doctors efficiently analyze the spike discharge of EEG data and further assist patients in diagnosis and treatment.