Research Area:  Machine Learning
Epilepsy is a significant burden on our society till now, due to appropriate healthcare treatment, cost of therapy, the spontaneous and unpredictable occurrence of seizures. There is a need for a fast and integrated neural investigation process that could help epileptologist to determine and diagnose the patients as soon as possible. Electroencephalogram (EEG) has been commonly used to diagnose patients by investigating the brain-s electrical activity that might be related to epilepsy. The proposed framework consists of several algorithms of feature extraction (current maxima, lower threshold, and target point selection), pattern matching (segment and domain matching) and post-processing with power, energy features. Maxima, homogeneity, power, energy and physiological field features have been used in this proposed model. Moreover, specific brain regions (lobes) inside the brain, where the seizure occurs, has been identified by the domain matching algorithm. There exist no such seizure detection system which provides warning message from the pre-ictal phase. This proposed model can be efficiently used as a real-time patient monitoring system which can send a warning message to the patient before the occurrence of seizure. This ultimately helps doctors for taking necessary actions. True positive rate (TPR) of 91.07% and 97.36% has been recorded for seizure and normal classes respectively. The accuracy and F1 score of the proposed model are 92.66% and 94.86%, respectively.
Author(s) Name:  Khakon Das, Debashis Daschakladar, Partha Pratim Roy, Atri Chatterjee, Shankar Prasad Saha
Journal name:  Biomedical Signal Processing and Control
Publisher name:  Elsevier
Volume Information:  Volume 57, March 2020, 101720
Paper Link:   https://www.sciencedirect.com/science/article/abs/pii/S1746809419303015