Patient-Specific Epileptic Seizure Onset Detection via Fused Eeg and Ecg Signals
MetadataShow full item record
Epilepsy is a neurological disorder that is associated with sudden and recurrent seizures. Epilepsy aﬀects 65 million people world-wide and is the third most common neurological disorder, after stroke and Alzheimer disease. During an epileptic seizure, the brain endures a transient period of abnormally excessive synchronous activity, leading to a state of havoc for many epileptic patients. Seizures can range from being mild and unnoticeable to extremely violent and life threating. Many epileptic individuals are not able to control their seizures with any form of treatment or therapy. These individuals often experience serious risk of injury, limited independence and mobility, and social isolation. In an attempt to increase the quality of life of epileptic individuals, much research has been dedicated to developing seizure onset detection systems that are capable of accurately and rapidly detecting signs of seizures. This thesis presents a novel seizure onset detection system that is based on the fusion of independent electroencephalogram (EEG) and electrocardiogram (ECG) based decisions. The EEG-based detector relies on a on a common spatial pattern (CSP)-based feature enhancement stage that enables better discrimination between seizure and non-seizure features. The EEG-based detector also introduces a novel classiﬁcation system that uses logical operators to pool support vector machine (SVM) seizure onset detections made independently across diﬀerent relevant EEG spectral bands. In the ECG-based detector, heart rate variability (HRV) is extracted and analyzed using a Matching-Pursuit and Wigner-Ville Distribution algorithm in order to eﬀectively extract meaningful HRV features representative of seizure and non-seizure states. Two fusion systems are adopted to fuse the EEG- and ECG-based decisions. In the ﬁrst system, EEG- and ECG-based decisions are directly fused to obtain a ﬁnal decision. The second fusion system adopts an over-ride option that allows for the EEG-based decision to over-ride the fusion-based decision in an event that the detector observes a string of EEG-based seizure decisions. The proposed detectors exhibit an improved performance, with respect to sensitivity and detection latency, compared with the state-of-the-art detectors. Experimental results demonstrate that the second detector achieves a sensitivity of 100%, detection latency of 2.6 seconds, and a speciﬁcity of 99.91% for the MAJORITY fusion case. In addition, a novel method to calculate the amount of neural synchrony that exists between the channels of an EEG matrix is carried out. This method is based on extracting the condition number from multi-channel EEG at a particular time instant to indicate the level of neural synchrony at that particular time instant. The proposed method of neural synchrony calculation is implemented in two detection systems. The ﬁrst system uses only neural synchrony as the feature for seizure classiﬁcation whereas the second system fuses energy and synchrony based decision to make a ﬁnal classiﬁcation decision. Both systems show promising results when tested on a set of clinical patients.
Qaraqe, Marwa Khalid (2016). Patient-Specific Epileptic Seizure Onset Detection via Fused Eeg and Ecg Signals. Doctoral dissertation, Texas A & M University. Available electronically from