Stacking ensemble based deep neural networks modeling for effective epileptic seizure detection

Highlights

• Epileptic seizure detection models were developed.

• The effectiveness of stacking ensemble approach was examined.

• All models were designed in Python by utilizing ‘Keras’ library with Tensorflow.

• Stacking ensemble-based model with 97.17% accuracy outperformed the baseline model.

Abstract

Electroencephalography signals obtained from the brain‘s electrical activity are commonly used for the diagnosis of neurological diseases. These signals indicate the electrical activity in the brain and contain information about the brain. Epilepsy, one of the most important diseases in the brain, manifests itself as a result of abnormal pathological oscillating activity of a group of neurons in the brain. Automated systems that employed the electroencephalography signals are being developed for the assessment and diagnosis of epileptic seizures. The aim of this study is to focus on the effectiveness of stacking ensemble approach based model for predicting whether there is epileptic seizure or not. So, this study enables the readers and researchers to examine the proposed stacking ensemble model. The benchmark clinical dataset provided by Bonn University was used to assess the proposed model. Comparative experiments were conducted by utilizing the proposed model and the base deep neural networks model to show the effectiveness of the proposed model for seizure detection. Experiments show that the proposed model is proven to be competitive to base DNN model. The results indicate that the performance of the epileptic seizure detection by the stacking ensemble based deep neural networks model is high; especially the average accuracy value of 97.17%. Also, its average sensitivity with 93.11% is superior to the base DNN model. Thus, it can be said that the proposed model can be included in an expert system or decision support system. In this context, this system would be precious for the clinical diagnosis and treatment of epilepsy.

Introduction

Epilepsy, which is one of the brain diseases, indicates itself as a blackout, out of balance physique acts, preternatural sensuality or muscle contraction as a result of the abnormal pathological oscillatory activity of a bunch of connected neurons in the brain (Kaya, Uyar, Tekin & Yıldırım, 2014; Raghu, Sriraam, Hegde & Kubben, 2019; Yuan, Zhou, Li & Cai, 2011; Zazzaro et al., 2019). This disease can cause physical injury and mental trauma or in the worst case scenario death because of reasons such as genetic and physical brain damage during seizures (Kalbkhani & Shayesteh, 2017; Kocadagli & Langari, 2017).

Epileptic seizures affect more than 50 million people in the world, but only 2/3 of them could be cure by antiepileptic drugs, and another 7–8% could be treated by surgical operations. Overall, 25% of individuals with epilepsy suffers because of lack of available therapy (Litt & Echauz, 2002).

Neurologists or field specialists mostly use Electroencephalography, in short called as EEG, signals to diagnose this disease (Rosas-Romero et al., 2019; Zazzaro et al., 2019). The EEG signals plotted by brain waves indicate the electrical activity of the brain and contain useful information about the condition of the brain Niedermeyer and Lopes da Silva (2005). The identification of these signals is conventionally performed by field-specialist. Manual scoring is liable to subject to human mistakes and is time-consuming, costly, and insufficient for creditworthy information (Bajaj & Pachori, 2012; Kabir, Siuly & Zhang, 2016; Kutlu, Kuntalp & Kuntalp, 2009). Therefore, automatic EEG seizure detection is of great importance for researchers in neuro-informatics. In other words, the diagnosis of this disease is a substantial subject in biomedical scrutiny and its evaluation. Automated systems are being developed to assess and detect epileptic seizure via EEG signals to prevent the situation which is field-specialist missing information (Kabir et al., 2016). The patients’ safety and life quality are improved if seizures can be predicted as early as possible before their occurrence (Chu, Chung, Jeong & Cho, 2017).

In sum, epilepsy detecting is conducted frequently by utilizing efficient automatic methods. Since the interpreting of EEG brain signals occupies many hours of neurologists, and reduces their efficiency, an expert system or decision support system is needed. In the literature, it has been observed that the studies that provide prediction performances of the base algorithms have been performed generally for EEG signals detection.

Due to the nature of science, new approaches are proposed day by day in terms of machine learning. In recent years, researchers have tried to find a more successful solution to improve the prediction performance in machine learning area. Inspired by this, in this study, a different methodological solution from the studies in literature that are of the same nature was employed to unleash an effective model with the utmost accuracy for automatically identification the two classes of EEG signals. The solution mainly focuses on the SEA-based modeling to improve the prediction performance. Stacking ensemble is a powerful technique which can enhance the prediction performance, and the model based on this approach is an up-to-date popular research topic in this field. It is core component of this study, and the fact that applying this model to epileptic seizure detection problem is the original side of this study. To the best my knowledge, so far no study has used SEA-based DNN modeling for epilepsy detection. To evaluate the performance of the model designed, 10-fold cross-validation technique was performed on training, validation and testing sets. The base DNN model was used also for evaluating and comparing of the performance of Stacking ensemble based DNN. In this context, the main focus of the study is to show the effectiveness of stacking ensemble approach where there is DNN as base learner and meta learner in automatically detection of epileptic seizure, and shows its performance better than base DNN model. Moreover, the main contributions of this study could be outlined as follows:

• Proposing of the SEA-based model to predict the epileptic seizure or not efficiently. This approach is valuable in terms of in expert and intelligent systems, and the experimental results are state-of-the-art.

• Adapting of the base DNN and SEA based DNN models, and comparing of them considering the accuracy, sensitivity and specificity metrics.

• Empirically results show that proposed model achieves superior performance to base DNN model; specifically, sensitivity measure.

The rest of the study is organized as follows. Section 2 presents the literature review. Section 3 describes the methodology in detail. Section 4 introduces the experimental design. Section 5 presents the experiments and results. Section 6 discusses the results and compares them with those by the state-of-the-art studies. Finally, Section 7 concludes the study.