Onset Detection of Epileptic Seizures From Accelerometry Signal Shitanshu Kusmakar, Chandan K. Karmakar, Bernard Yan, Terence J.O’Brien, Ramanathan Muthuganapathy and Marimuthu Palaniswami Abstract— Epileptic seizures are the result of any abnormal asynchronous firing of cortical neurons. Seizures are abrupt and pose a risk of injury and fatal harm to the patient. Epilepsy affects patients quality of life (QOL) and imposes financial, social, and physical burden on the patient. The unpredictability associated with seizures further adds to the reduced QOL and increases dependence on caregivers and family members. A seizure triggered alarm system can reduce the risk of seizure- related injuries and aid in improving patient’s QOL. This study presents real-time onset detection of seizures from accelerome- try signal. An automated approach based on statistical machine learning is employed to learn the onset of seizures. To search for the optimal parameter that simultaneously maximizes detection sensitivity while minimizing false alarms and latency, the epoch length is varied from t = {1,..., 10s}. Linear and non-linear time-varying dynamical patterns were extracted from every epoch using Poincar´ e plot analysis. The correlation patterns were learned using a kernalized support vector data descriptor. The preliminary analysis on accelerometry data collected from 8 epileptic patients with 9 generalized tonic-clonic seizures (GTCS) shows promising results. The proposed algorithm detected all GTCS events (sens: 100%, FAR: 1.09/24h) at 8s from onset. The proposed algorithm can lead to a sensitive, specific, and a relatively short-latency detection system for real- time remote monitoring of epileptic patients. I. INTRODUCTION Epilepsy affects approximately 1% of the global popula- tion and approximately 50 million people are currently living with epilepsy [1]. Despite of the rapid advances in drug dis- covery and alternate treatment strategies for epilepsy, a large number of patients continue to have seizures. Patients with poor control of seizures lead a poor quality of life (QOL) due to passive coping styles, social consequences, and level of independence [2]. In addition, patients with poor control of seizures are more prone to epilepsy associated morbidity and mortality. Recent research, indicates that appropriate intervention following a seizure can reduce the risk of injury and harm [3]. In this study, we describe a framework of an algorithm for real-time onset detection of generalized tonic- clonic seizures (GTCS) from accelerometry signal. The pro- posed technique can provide a wearable remote monitoring seizure triggered alarm system. S. Kusmakar and M. Palaniswami are with the Department of Electrical and Electronic Engineering, The University of Melbourne, Vic - 3052, Aus- tralia. skusmakar@student.unimelb.edu.au, palani@unimelb.edu.au C. K. Karmakar is with Department of Electrical and Electronic Engineer- ing, University of Melbourne, Vic - 3052, Australia and also with Deakin University Geelong, Vic - 3125, Australia. karmakar@unimelb.edu.au B. Yan and T. O’Brien are with the Melbourne Brain Centre, Royal Melbourne Hospital, Dept. of Medicine, The University of Melbourne, Vic - 3052, Australia. Bernard.Yan@mh.org.au, obrientj@unimelb.edu.au R. Muthuganapathy is with the Department of Engineering Design, Indian Institute of Technology Madras, India. mraman@iitm.ac.in Accelerometers have been previously used in automated detection of GTCS events however, most systems suffer from high FAR or high latencies or both [4]. Several groups have also investigated the use of multi-modality systems to improve the specificity of the seizure detection system [5] [6]. However, these systems are optimized to detect the clonic phase of GTCS events and thus, result in a higher detection latency. Conradsen et. al. [7] proposed an algorithm for onset detection of GTCS event using surface electromyography (sEMG) zero-crossing rate. They detected all GTCS events from 11 patients with a mean detection latency of 13.7s while, the false alarm rate (FAR) was 1/24h. Other approaches on onset detection of seizures are based on utilizing electroencephalography (EEG) or intracranial EEG (iEEG) recordings [8]. The best performing system are based on iEEG (sens: 97%, FAR: 0.6/24h, latency: 5s) [9] or EEG with 60 scalp electrodes (sens: 100%; FAR: 0.55/24h; latency: 4s) [10]. However, iEEG is an invasive procedure and is associated with a high risk of infection while, systems based on sEMG and EEG are susceptible to movement artifacts and can be uncomfortable during continuous use [4]. In this study, we presents a generic non-patient spe- cific algorithm for real-time onset detection of tonic-clonic seizures using wrist-worn accelerometer based device. The algorithm detects seizures by examining the correlation pat- terns derived from short-length signals characterizing the onset of events. The developed algorithm was tested on ac- celerometer data recorded from 8 patients undergoing video- electroencephalography monitoring (VEM). The preliminary results showed a higher seizure detection sensitivity and a lower detection latency in comparison to the state-of-art non- EEG ambulatory monitoring systems. II. METHODS The onset detection of seizures is framed as a binary classification problem that involves differentiation of seizure activity from non-seizure activity. Continuous monitoring accelerometer devices record a vast amount of daily living activities. Therefore, an algorithm was developed in the first stage to discard no movement or subtle movement activities. In the second stage, the epoch’s are classified into seizure and non-seizure activities using Poincar´ e derived descriptors and kernalized support vector data description (SVDD) classifier. The following sections presents the details of the two stage seizure detection algorithm.