Research And Implementation Of Key Technologies For Wearable Seizure Monitoring System

Wearable Seizure Monitoring System(WSMS)is of great significance for adjuvant treatment and health management of epilepsy.At present,WSMS is mainly implemented by using non-electroencephalogram(non-EEG)signals(e.g.acceleration and muscle activity)to detect seizures,whose application range is limited to the detection of convulsive seizure.As the key biological signal to capture the electrical activity in cerebral cortex,EEG provides the necessary information for seizure detection,and can be employed to detect almost all the types of seizure,including non-convulsive seizure.Therefore,EEG based WSMS has become a research hotspot in this field,but remains a challenging topic in meeting clinical requirements in terms of detection performance,power consumption and portability.To resolve the problem mentioned above,this thesis investigates the key technologies involved in the implementation of EEG based WSMS,including the design of seizure detection algorithm,hardware implementation and optimization based on FPGA,the development of Android client software for epilepsy management and the building of verification platform.The main contributions are listed as follows:(1)For the problem of missed detection rate caused by unbalanced characteristics of epileptiform EEG data,the balanced algorithm based on K-means is used to cluster samples from the majority class,which alleviates the problem of mistaken deletion of samples from class boundary and reduces the imbalance rate effectively.The effectiveness of imbalanced processing method is validated by the experiment using CHB-MIT EEG dataset.(2)Aimed at the limitations of the Generalized Hurst Exponent(GHE)method in describing the state of EEG signal fluctuations within the time window,this thesis proposes a feature extraction method combining Exponential Moving Average(EMA)with GHE(i.e.EMA-GHE)to calculate the long-term correlation characteristics of epileptiform EEG signals.Also,Random Forest(RF)is employed as the classifier.Then,the seizure detection algorithm suitable for WSMS can be obtained.In order to verify the robustness of proposed algorithm,the fragment-based evaluation standard is used.The experiment results show that average detection rate and false detection rate are98.89% and 1.65%,respectively.To show the advantages of proposed algorithm in implementing seizure detection with high detection accuracy and low latency,theevent-based evaluation standard is also used.The experiment results show that the average detection rate,false detection rate and detection delay are 98.18%,0.0259times/hour and 0.32 seconds,respectively.(3)The method of hardware implementation and optimization for seizure detection algorithm based on FPGA is proposed.And the IP core of seizure detection is designed on Vivado HLS including data bit width selection,EMA-GHE feature extraction,and RF model prediction.According to the parallel characteristics of proposed algorithm,the multiplexing strategy and pipeline design are used to optimize the IP core.Experiment result shows that optimization method of IP core on Vivado HLS can realize a "perfect" mapping from algorithm design to hardware implementation.(4)The WSMS verification platform is built,mainly including three modules,i.e.epileptiform EEG data transmission,seizure detection and Android client feedback.The experiment result shows that the verification platform can realize the function of seizure monitoring.