LAUSR

Epilepsy is a commonly observed long-term neurological disorder that impairs nerve cell activity in the brain and has a severe impact on people’s daily lives. Accurate seizure detection in the long-term electroencephalogram (EEG) signals has gained vital importance in the diagnosis of patients with epilepsy. Visual interpretation and detection of epileptic seizures in long-term EEG is a time-consuming and burdensome task for neurologists. Therefore, in this study, we propose a computationally efficient automated seizure detection model using a novel feature called successive decomposition index (SDI). We observed that the SDI feature was significantly higher during the epileptic seizure as compared to normal EEG. The performance of the proposed method was evaluated using three databases, namely the Ramaiah Medical College and Hospital (DB1), CHB-MIT (DB2) and the Temple University Hospital (DB3) consisting of 58 h, 884 h, and 408 h of EEG, respectively. Experimental results revealed the sensitivity–false detection rate–median detection delay of 97.53%–0.4/h–1.5 s, 97.28%–0.57/h–1.7 s, and 95.80%–0.49/h–1.5 s for DB1, DB2, and DB3, respectively, using the support vector machine classifier. The proposed method significantly outperformed previously presented methods (wavelets and other feature extraction methods) while being computationally more efficient. Further, to the best of the author’s knowledge, present study is the first study that was tested on three different EEG databases and showed consistent results leading to the generalization and robustness of the algorithm. Hence, the proposed method is an efficient tool for neurologists to detect epileptic seizures in long-term EEG.