Detecting the onset of an epileptic seizure using a novel time-series approach
MetadataShow full item record
Electroencephalography (EEG) is one of the most popular non-invasive techniques for acquiring electrical signals from the brain. Data mining EEG signals finds numerous applications in the field of neuroscience for obtaining crucial information about the neural activities. EEG data is very complex in that it is non-stationary and multidimensional. Therefore, the task is how to convert voluminous raw EEG data into a succinct representation. This research provides a methodology for representing EEG data in a concise and comprehensible format using a minimum number of data points without the loss of useful information. Among many applications of studying EEG data, detecting epileptic seizures concerns neurologists the most. Epilepsy is a serious disorder characterized by the occurrence of epileptic seizures. These seizures occur as a result of abnormal neuronal activities of the brain. Today, more than 65 million people in the world suffer from epileptic seizures which can be life-threatening. It is not just the physical effects of seizure that impacts patients adversely but also the social isolation that the patient and their families face. If EEG signals are analyzed properly, seizures can be predicted at their onset. This thesis proposes a seizure prediction method which uses a novel time-series approach to provide a useful method for the diagnosis of epileptic seizures. The key to the method identifies transitions from non-epileptic (pre-ictal) to epileptic (ictal) segments of the EEG signal using offset statistical moving averages. This research examines EEG data of multiple epileptic patients from CHB MIT database. The method analyzes EEG signals for common transitional patterns using multiple inter-patient and intra-patient seizure files. The experiments provide substantial results and predict seizures early in some situations and with a minimal latency in a few other situations.
Institutional Repository URIhttp://hdl.handle.net/10657.1/1034