Research On Embedded Automatic Detection System For Epilepsy

Electroencephalography(EEG)plays an important role in the diagnosis and treatment of epilepsy.Traditional epilepsy diagnosis requires professional doctors to search for seizure data from a large number of EEGs based on clinical experience.This is a time-consuming and laborious process,and the results greatly depend on the subjective judgment of neurologists,prone to misdiagnosis or missed diagnosis.Therefore,the research of automatic seizure detection system based on EEG can not only reduce the burden of clinical diagnosis for doctors,but also help to improve the accuracy and objectivity of diagnosis.The normal automatic seizure detection method is the combination of a feature extractor and a classifier.The feature extraction aims to obtain the representative characteristics of data,including time domain analysis,frequency domain analysis using FFT,wavelet transform,empirical mode decomposition,and nonlinear dynamics analysis,etc.Classifiers mainly include Bayes,K nearest neighbor,design tree,support vector machine,neural network,etc.which can distinguish EEG signals according to their features.This paper aims at clinical automatic detection of seizures,and three basic models are proposed based on time-frequency characteristics of EEG and support vector machine with,one-dimensional convolutional neural network and Two-dimensional convolutional neural network.Because of the difference between the EEG signals of different patients and using transfer learning as reference,the EEG data of multiple patients were used to form the general dataset to train the general model of the above three algorithms.The remaining EEG data of each patient were used to form personalized dataset to train the personalized model of each patients based on the general model.A complete data processing flow of "data acquisitiondata reception-pre-processing-algorithm recognition-post-processing-visual display" was constructed on the Raspberry Pi 4B platform.Experiments were conducted on the Bonn dataset and CHB-MIT dataset to verify the accuracy and real-time performance of the algorithms.The embedded automatic seizure detection system built in this paper has the characteristics of low cost,high accuracy and real-time detection,which can meet the performance requirements of embedded seizure detection.