Abstract—Penicillin-induced focal epilepsy is a well-known model in epilepsy research. In this model, epileptic activity is generated by delivering penicillin focally to the cortex. The drug induces interictal electroencephalographic (EEG) spikes which evolve in time and may later change to ictal discharges. This paper proposes a method for automatic classification of these interictal epileptic spikes using iterative K-means clustering. The method is shown to be able to detect different spike waveforms and describe their characteristic occurrence in time during penicillin-induced focal epilepsy. The study offers potential for future research by providing a method to objectively and quantitatively analyze the time sequence of interictal epileptic activity. I. INTRODUCTION PILEPSY is a chronic neurological disorder characterized by recurrent interruptions of normal brain function called epileptic seizures [1]. These are abnormal hypersynchronous electrical discharges of the brain accompanied by behavioral manifestations such as altered awareness, generalized tonic–clonic movements, and manual automatisms [2]. About 50 million people worldwide suffer from epilepsy [3]. Even though significant progression has lately been made in the brain imaging technologies, electroencephalogram (EEG) still has an essential role in the diagnosis of epilepsy. Characteristic EEG changes, such as generalized spike-wave discharges, can be seen during epileptic seizures [4]. In addition to ictal activity, patients with epilepsy usually have abnormal EEG changes also between seizures. These interictal epileptiform discharges (IED) can be divided morphologically into sharp waves, spikes, spike-wave complexes, and polyspike-wave complexes [5]. IEDs are considered to contain valuable information related to the Manuscript received June 22, 2010. This work was supported in part by the GETA Graduate School, Walter Ahlström Foundation, Tauno Tönning Foundation, Finnish Foundation for Economic and Technology Sciences – KAUTE, The Finnish Medical Foundation, and Orion-Farmos Research Foundation. J. Kortelainen is with the Department of Electrical and Information Engineering, BOX 4500, FIN-90014 University of Oulu, Finland (e-mail: jukka.kortelainen@ee.oulu.fi). M. Silfverhuth and T. Seppänen are with the Department of Electrical and Information Engineering, University of Oulu, Finland. K. Suominen is with the Department of Clinical Neurophysiology, Oulu University Hospital, Finland. E. Sonkajärvi and S. Alahuhta are with the Department of Anesthesiology, University of Oulu, Finland. V. Jäntti is with the Department of Biomedical Engineering, Tampere University of Technology and Department of Clinical Neurophysiology, Seinäjoki Central Hospital, Finland. diagnosis, treatment, and prognosis of the disease [6], [7]. Penicillin-induced focal epilepsy is a well-known model in the epilepsy research. In this model, epileptic activity is generated in the brain of an animal by delivering penicillin focally to the cortex. The drug prevents GABA-mediated inhibitory control of the main population of pyramidal neurons by blocking the GABA A receptors [8], [9]. The impaired inhibition induces epileptic EEG spikes which evolve in time and may later change to ictal discharges [10]. Experimental focal epilepsy is widely used to dissect various cellular and synaptic processes and thereby to explain the mechanisms of epileptic activity [9]. In this paper, a method for the classification of EEG spikes during induced focal epilepsy is proposed. The purpose is to be able to detect automatically different spike waveforms and describe their characteristic occurrence in time after delivering penicillin to the cortex. The proposed method is tested with data from a previous experimental EEG/fMRI series of a focal epilepsy model [11]. The clustering algorithm, spike classification, and data acquisition procedure are explained in Section II. Section III presents the results. In Section IV, the paper is concluded and a short discussion about the results and future work is given. II. MATERIALS AND METHODS A. Clustering Algorithm and Spike Classification The clustering algorithm and spike classification were implemented using the Matlab technical computing language (The MathWorks Inc., Natick, MA). A schematic representation of the clustering algorithm is given in Fig. 1. The algorithm is based on iterative K-means clustering and can be divided into four parts: 1) Sequence extraction 2) Spike extraction 3) K-means clustering 4) Centroid comparison In sequence extraction, the data sequences containing an epileptic spike are extracted from the EEG recording. The spike positions are roughly determined by amplitude thresholding. A proper threshold value is approximated using visual inspection of the EEG amplitude changes after penicillin administration and the position of the first epileptic spike. The data sequences of predefined length are then extracted around the positions in which this threshold value Automatic Classification of Penicillin-induced Epileptic EEG Spikes Jukka Kortelainen, Student Member, IEEE, Minna Silfverhuth, Kalervo Suominen, Eila Sonkajärvi, Seppo Alahuhta, Ville Jäntti, and Tapio Seppänen E 32nd Annual International Conference of the IEEE EMBS Buenos Aires, Argentina, August 31 - September 4, 2010 978-1-4244-4124-2/10/$25.00 ©2010 IEEE 6674