Automatic Seizure Detection Using Stockwell Transform And Boosting Algorithm

Epilepsy is one kind chronic disorder that results from excessive and hypersynchronous discharges in the brain neurons, which leading to the disturbances of brain function. The clinical features of epilepsy seizures contains unexpectedness, repeatability, transient and so on. Epilepsy is one of the common disease in the nervous system, whose reduplicative, epileptic seizures bring great pains to the patients and seriously affect their living quality. Electroencephalogram records the electrical activities of the brain by electrode installed on the cerebral cortex. As epilepsy seizures attributed to excessive abnormal discharges of certain nerve cells in the brain, so EEG becomes the most important tool in epilepsy diagnosis and treatment. EEG contains the physiological and pathological information that can’t be replaced by other detection methods. At present, EEG recordings are visually checked by trained neurologists, and it is a heavy and time-consuming task due to the massive amounts of EEG data. Therefore, automatic seizure detection is particularly necessary to reduce the burden of neurologists and improve the neurologists and improve the diagnosis efficiency.During epilepsy seizures, nerve cells in the brain generate various waveforms in rhythmic, including spike wave, SWDs, slow wave, fast wave and so on, and its rangeability of waveforms are more apparent than intermission. Epilepsy detection extracts significant features that represent different periods for classification and identification, so as to achieve the purpose of automatic seizure detection.In this paper, we proposed a novel automatic seizure detection methods on the basis of predecessors’ research. This method based on Stockwell transform and Gradient boosting algorithm and compared with other techniques. Firstly, after preprocessing EEG signals, we applied Stockwell transform and mapped the time-frequency characteristics on the time-frequency plane; Secondly, the time-frequency characteristics plane was segmented into 12 parts according to certain rules and extracted the power spectral density of each parts to quantitatively describe the brain activities. Thirdly, the extracted feature vectors were transmitted into classifier for training and classification; Finally, post-processing was applied on the raw results and then the automatic seizure detection was realized.In this experiment, the EEG data comes from the Freiburg EEG database, whose database contains 21 patients of 87 epileptic seizures. After comparatively analyzing with other methods, the proposed method in our paper can effectively identify epileptic seizures, and it has the advantages of faster computation and stronger real-time monitoring, which is more conductive to clinical application.