Non-parametric early seizure detection in an animal model of temporal lobe epilepsy

Early seizure detection in an animal model of temporal lobe epilepsy Sachin S. Talathi∗ , Dong-Uk Hwang∗ , William Ditto∗ and Paul R Carney∗,† ∗

J Crayton Pruitt Family Department of Biomedical Engineering, University of Florida, FL 32611 † Department of Pediatrics, Division of Neurology,Department of Neurosceince,University of Florida, FL 32611

Abstract. The performance of five seizure detection schemes, i.e., Nonlinear embedding delay, Hurst scaling, Wavelet Scale, autocorrelation and gradient of accumulated energy, in their ability to detect EEG seizures close to the seizure onset time were evaluated to determine the feasibility of their application in the development of a real time closed loop seizure intervention program (RCLSIP). The criteria chosen for the performance evaluation were, high statistical robustness as determined through the predictability index, the sensitivity and the specificity of a given measure to detect an EEG seizure, the lag in seizure detection with respect to the EEG seizure onset time, as determined through visual inspection and the computational efficiency for each detection measure. An optimality function was designed to evaluate the overall performance of each measure dependent on the criteria chosen. While each of the above measures analyzed for seizure detection performed very well in terms of the statistical parameters, the nonlinear embedding delay measure was found to have the highest optimality index due to its ability to detect seizure very close to the EEG seizure onset time, thereby making it the most suitable dynamical measure in the development of RCLSIP in rat model with chronic limbic epilepsy. Keywords: Epilepsy, Rat model, EEG, Seizure detection PACS: 87.80.Tq, 87.80.Xa

1. INTRODUCTION Epilepsy is a syndrome of episodic brain dysfunction characterized by recurrent seemingly unpredictable spontaneous seizures [1]. The occurrence of the seizure in patients without any forewarning is the most debilitating aspect of the disease. A lot of scientific research has therefore focused on developing methods for predicting, anticipating, forecasting or detecting the seizure early enough [2] such that it could then facilitate timely therapeutic intervention and thereby improve the life of an epileptic patient. The final goal of any seizure prediction or detection scheme is in its application in development of system that not only forewarns the occurrence of seizure but also takes measures to prevent the impending seizure from happening. A survey of the current seizure prediction tools by [2] suggest that many of these algorithms have not been subject to rigorous statistical evaluation in terms of their predictive power (defined as the ration of true positive to the sum of true positive and false positives ), sensitivity (defined as ratio of true positive to the sum of true positive and false negative), and specificity (defined as the ratio of true negative to sum of true negative and false positive) to warrent a closed-loop seizure prevention study. The authors suggest that in order to advance towards a clinical application, future studies on seizure forecasting or prediction should place a strong emphasis on a sound methodology

and include rigorous statistical validation. The absence of sound methodology in the development of a robust seizure prediction algorithm is intrinsically linked to the inherent variability in the EEG data recorded from epileptic patients. Possible reasons for such high variability are the confounding effects of epilepsy types and severity in patients, the pathology of the disease, the effects of anticonvulsant drug treatments, the recording sites and the over all environment surrounding patient care. Animal models of epilepsy that emulate human epileptic condition provide an opportunity to generate experimental EEG data under controlled experimental conditions, thereby minimizing the effect of confounding variables that affect the performance of seizure prediction algorithms. The EEG data generated from these animal models can then form a test bed to evaluate the performance of any given seizure prediction algorithm. While seizure prediction algorithms aim to detect pre-ictal state minutes to hours in advance of an impending seizure, early seizure detection algorithms aim to identify the seizure onset zone in the EEG data, which may occur few seconds in advance of the clinical manifestation of the seizures. Statistical robustness of early seizure detection schemes [3, 4, 5] make them more attractive for use in the development of closed loop intervention system [6]. However there is still a lack of any comparative study for seizure detection methods in terms of their performance for use in real-time closed loop intervention system. In this chapter we present a comparative study on a number of different univariate measures derived from the EEG data for seizure detection. We used EEG data from rat model of chronic limbic epilepsy (CLE) in order to compare different dynamical measures for seizure detection which can be utilized in the development of a closed loop seizure intervention system. The following parameters are compared to determine the most efficient measure for seizure detection : (a) Sensitivity (b) Specificity (c) Predictability (d) Computational efficiency and (e) The mean delay in the seizure detection by the algorithm with respect to electrographic seizure onset as determined by visual inspection. The chapter is organized as follows. In Section 2, we first describe the experimental procedure for preparing the CLE rat model. We then present the mathematical details on the number of measures studied in this chapter. We then describe the statistical approach used to determine the sensitivity and specificity for each measure used for seizure detection and present an empirical optimality function that is used to compare various measures for seizure detection. In Section 3, we begin with the demonstration of various EEG seizure morphology and the common features shared by each EEG seizure. We then present the results derived from each of the measure used for seizure detection. Table 2, summarizes the final results.

2. MATERIAL AND METHODS 2.1. Experimental procedure for developing rat model for chronic limbic epilepsy 2.1.1. Animals Experiments were performed on 2-month old male Sprague Dawley rats (n= 9) weighing 210-265 g. Protocols and procedures were approved by the Institutional Animal Care and Use Committee of the University of Florida.

2.1.2. Surgery and electrode implantation The top of the head was shaved and chemically sterilized with iodine. The skull was exposed by a midsagittal incision that began between the eyes and extended caudally to the level of the ears to expose the bregma and lambdoidal suture. A peroxide wash was applied to the skull to remove excess soft tissue. Four 0.8 mm stainless steel screws (Small Parts, Miami Lakes, FL) were placed in the skull to anchor the acrylic headset: 1) two screws were AP 2 mm and bilaterally 2 mm; 2) one screw was AP -3 mm and left 2 mm and served as a ground electrode; and 3) one screw was AP -2 mm to the lambdoidal suture and right 2 mm and served as a reference electrode. Holes were drilled to permit insertion of 2 stainless steel bipolar twist electrodes (1 mm tip separation) into the left and right ventral hippocampi for electrical stimulation. In 3 of the 9 CLE rats, in addition holes were drilled to insert 2 recording stainless steel electrodes (AP -5.3 mm, bilateral 4.9 mm, vertical -5 mm below the dura) in the left and the right ventral hippocampi. 2 stainless steel monopolar recording electrodes were also placed in the bilateral frontal cortices (AP 3.2 mm, bilateral 1 mm, vertical -2.5 mm below the dura). Electrodes were labeled according to their relative positions on the ratÕs skull as LF/RF (left/right frontal cortex) and LH/RH (left/right hippocampus). Electrode pins were positioned in a plastic strip connector and the entire headset was glued into place using cranioplast cement (Plastics One, Inc., Roanoke, VA). In 6 of the 9 rats with CLE, recordings were done with sixteen microwire recording electrodes (50 µm polyimide insulated tungsten microwires) were implanted to the left of the midline suture in a diagonal fashion in the CA1, CA2, CA3 and the dentate gyrus region of the hippocampus. The orientation of the microelectrodes was such that the furthest right microwire was positioned at 4mm caudal to bregma, 1.7mm to the left of midline suture and at a depth of 4mm from the midline suture. The microelectrodes were configured in four bundles (four microelectrodes in each bundle) arranged in rectangular pattern to conform with the morphology of the hippocampus. On the long axis of the bundle each microelectrode was seperated by 200µm and on the short axis, the separation was 400µm. A second array of sixteen microelectordes was place to the right of the midline in the same orientation. The two seizing rats also contained one bipolar twisted Teflon sheathed stainless steel electrode of 330µm diameter, implanted in the contralateral posterior ventral hippocampus (5.3 mm caudal to bregma, 4.9mm to the right of midline suture and at a depth of 5mm from

the cortical surface) for stimulating the rat into self sustaining status epilepticus. All the electrodes were chronically secured with dental cement. Following surgery, animals were allowed to recover for 1 week prior to additional procedures.

2.1.3. Induction of status epilepticus All animals were stimulated electrically 1 week after surgery to induce status epilepticus. During electrical stimulation and iEEG acquisition, animals were housed in specially-made chambers (Bertram et al., 1997). Stimulus trains were delivered for 5070 min with a duty cycle of 10 sec on/ 2 sec off, consisting of biphasic square wave pulses at a frequency of 50 Hz, a pulse duration of 1 ms, and intensities of 300Ð400 mA. During the stimulus, a normal behavioral response was for the animal to display Òwet dog shakesÓ and increased exploratory activity. After approximately 20-30 min of stimulation, convulsive seizures (up to 1 min in duration) were usually observed about every 10 min. At the end of the stimulus period, continuous iEEG recordings were observed for evidence of slow waves in all recorded channels. If slow waves were not demonstrated, then the stimulus was re-applied for 10 min intervals 1Ð3 times until continuous slow waves appeared following termination of the stimulus. Lack of response to this stimulation protocol was infrequent (