Channel Selection and Feature Enhancement for Improved Epileptic Seizure Onset Detector Marwa Qaraqe * , Muhammad Ismail † , Qammer Abbasi † , and Erchin Serpedin * * Department of Electrical and Computer Engineering Texas A&M University, College Station, TX, 77843–3128, USA Email: marwa@tamu.edu and serpedin@ece.tamu.edu † Department of Electrical and Computer Engineering Texas A&M University at Qatar, Education City, Doha, Qatar, 23874 Email: m.ismail@qatar.tamu.edu and qammer.abbasi@qatar.tamu.edu Abstract—This paper presents a novel architecture for a patient-specific epileptic seizure onset detector using scalp elec- troencephalography. The proposed architecture exploits the bene- fits of both channel selection and feature enhancement to improve the detector performance. The novel architecture results in higher energy difference between the pre-seizure and seizure states and hence performs better in terms of detection sensitivity and false alarm rate compared to benchmark detectors available in the literature. In detail, the proposed architecture achieves a 7% increase in sensitivity and a reduction of 9 false alarms per hour compared to the benchmark detector. Index Terms—epilepsy, EEG, seizure onset I. I NTRODUCTION Epilepsy is a brain disorder that is characterized by inter- mittent abnormal firing of neurons, called seizures. Neurons normally generate electrochemical impulses that act on other neurons, glands, and muscles to produce human thoughts, feelings, and actions. However, in epilepsy, the normal rhythm of neural firing is disturbed, causing the epileptic patient to experience strange sensations, emotions, and behaviors, or sometimes convulsions, muscle spasms, and loss of conscious- ness [1]. Epilepsy affects approximately 1% of the population in the United States. Around 80% of those patients can control their seizures with modern medicines and surgical techniques. However, nearly 25 to 30 percent of patients are diagnosed with intractable epilepsy, where they cannot control seizures even with the best available treatment [2]. The confusion, loss of consciousness, and lack of muscle control that accompany certain types of seizures can lead to serious injuries that include fractures, head injuries, and burns. These injuries account for a significant component of the risk associated with epilepsy [3]. The risk of injury associated with epilepsy can be mitigated by using a device that can reliably detect or predict the onset of seizure episodes. Because the clinical behavior of an epileptic seizure is preceded by and then accompanied by electroencephalographic alterations, electroencephalography (EEG) can be used to measure these alterations [4]. The scalp EEG is a non-invasive, multi-electrode recording of time-varying potentials generated by the neurons located on the cerebral cortex. The electrodes are distributed sym- metrically around the scalp to provide a temporal and spatial summary of the brain’s electrical activity. The EEG activity of clinical relevance is limited to the frequency band 0.5 -50 Hz, and that of seizure activity is further limited to the frequency band 0.5 - 25 Hz [5]. Extensive research has been dedicated to the detection of the earliest signs of electrographic changes associated with a seizure using either scalp or inter-cranial EEG. A device that has the ability to detect the electrographic onset of a seizure will enable epileptic patients to lead a more normal and secure life, and will help them to avoid injuries due to the sudden nature of the seizure. In [7], one of the earliest automated systems for the detection of epileptic activity in long-term EEG recordings is designed and implemented by applying empirically determined thresholds on time-domain features. A seizure onset detection algorithm that processes a sin- gle, manually-selected channel of an invasive-EEG recording (ECoG) is implemented in [8] by using a maximum-likelihood classifier with Gaussian mixture model conditional densities to differentiate between a patient’s normal and abnormal ECoG. In [5], a patient-specific method for the detection of epileptic seizure onset from scalp-EEG is designed using wavelet de- composition, feature extraction, and a support vector machine (SVM) classification algorithm. In [9], automatic detection of epileptic seizure event and onset is proposed using wavelet based features and certain statistical features without wavelet decomposition. In [10], different methods are employed for EEG chan- nel selection preceding automatic seizure detection. Channel selection based on the highest variance method has demon- strated an improved performance compared to a no channel selection scheme. EEG channel selection reduces the detector computational complexity and avoids using channels with no relevant information that may deteriorate the detector’s performance. In [11], it has been shown that EEG signal MOBIHEALTH 2014, November 03-05, Athens, Greece Copyright © 2014 ICST DOI 10.4108/icst.mobihealth.2014.257277