ORIGINAL PAPER Can Neural Network Able to Estimate the Prognosis of Epilepsy Patients Accorrding to Risk Factors? Kezban Aslan & Hacer Bozdemir & Cenk Sahin & S. Noyan Ogulata Received: 26 November 2008 / Accepted: 16 February 2009 / Published online: 28 March 2009 # Springer Science + Business Media, LLC 2009 Abstract The aim of this study is to evaluate the underlying etiologic factors of epilepsy patients and to predict the prognosis of these patients by using a Multi- Layer Perceptron Neural Network (MLPNN) according to risk factors. 758 patients with epilepsy diagnosis are included in this study. The MLPNNs were trained by the parameters of demographic properties of the patients and risk factors of the disease. The results show that the most crucial risk factor of the epilepsy patients was constituted by the febrile convulsion (21.9%), the kinship of parents (22.3%), the history of epileptic relatives (21.6%) and the history of head injury (18.6%). We had 91.1 % correct prediction rate for detection of the prognosis by using the MLPNN algorithm. The results indicate that the correct prediction rate of prognosis of the MLPNN model for epilepsy diseases is found satisfactory. Keywords Epilepsy prognosis . Risk factors . MLPNN . Levenberg-Marquardt. Introduction The studies on the prognosis of epilepsy have been difficult because of methodological problems and because epilepsy may have many different underlying etiologies [1, 2]. In addition, we do not know how much these etiological or risk factors affect the prognosis. The overall risk of recurrence after a first seizure is an important aspect of prognosis [1, 3]. For this reason, the estimation of the prognosis of epilepsy will undoubtedly influence the treatment strategy. Several aspects must be considered in epilepsy progno- sis; these include the likelihood of seizure recurrence after a first epileptic attack; the underlying etiology of this attack; the role of antiepileptic drug (AED) treatment in outcome; the risk of recurrence of seizures after AED discontinuation, etc [1–4]. However, uncertainties about the issues of the predictability of the remission (who will achieve remission and when?), and the role of treatment in outcome remain. Recent developments in this field show that the trend is to develop new methods for computer-assisted decision- making in medicine and to evaluate critically these methods in clinical practice. This is manifested by an increasing number of medical devices currently available on the market with embedded AI algorithms, together with an accelerating pace of publication in medical journals. One of these techniques is artificial neural networks (ANNs). ANNs have been used successfully and extensively in many different problems in medicine and for the detection of prognosis of different diseases such as the prognosis of breast cancer, epidermolysis bullosa simplex, nasopharyn- geal carcinoma, heart failure, melanoma etc [5–16]. ANNs have also been used for the detection of seizure activity [17–22]. However, there has been no study for the detection of prognosis of epilepsy patients using neural networks. That is why we try to predict the prognosis of epilepsy by using neural network in this study. This study aimed to define the prospects of newly diagnosed epilepsy, assess the dynamics of its course and J Med Syst (2010) 34:541–550 DOI 10.1007/s10916-009-9267-8 K. Aslan (*) : H. Bozdemir Department of Neurology, Faculty of Medicine, Cukurova University, Adana, Turkey e-mail: kezbanaslan@hotmail.com C. Sahin : S. Noyan Ogulata Department of Industrial Engineering, Faculty of Engineering and Architecture, Cukurova University, Adana, Turkey