Research On Epileptic Seizure Detection Based On Deep Learning

Epilepsy is a common chronic neurological disorder,it will outbreak randomly and repeatedly,causing a huge negative impact on the patient’s physical and mental health and normal life.Recognizing seizures quickly and accurately is of great significance to patients with epilepsy.It can not only guide patients to take medication reasonably,but also enable the family of patients to be informed of the condition in time,so that patients in seizures can be rescued.At present,the diagnosis of epilepsy and the recording and identification of seizures are mainly determined by medical personnel performing electroencephalogram(EEG)examination and analysis on patients.Although this method is accurate,the process of obtaining and analyzing the patient’s EEG is very time-consuming and laborious,and cannot meet the needs of patients to monitor their physical conditions anytime,anywhere.Therefore,automatic and convenient monitoring of the condition of epilepsy patients and more accurate identification of seizures are urgent needs to track the condition of epilepsy patients and the focus of future research.With this as the background,we used a portable sensor device and developed a corresponding deep learning algorithm to study the recognition of seizures.We combined with the recently developed wearable health monitoring equipment,through the acceleration sensor installed on a wristband bracelet platform,to obtain the wearer’s limb movement signal,that is,the wrist’s triaxial acceleration signal,provides the original data for the recognition of epileptic seizures,and at the same time solved the problem of portability of traditional EEG acquisition devices.In order to analyze more effectively,we have preprocessed the collected raw data,including filtering and noise reduction,window segmentation,standardization,and time-frequency analysis.Three neural network algorithm models based on deep learning were constructed,including fully connected neural network models,two-dimensional convolutional neural network models and long-short term memory neural network models.By comparing the recognition effects of the above different algorithm models,we found a suitable recognition algorithm,made appropriate improvements,improved the accuracy of recognition,reduced the false positive rate and false negative rate in the recognition process,and finally established a more mature seizure recognition algorithm based on deep learning and wrist vibration acceleration.