Improving the Performance of ECG-based Epileptic Seizure Prediction

Epileptic seizure is a neurological disorder affecting 50 million people worldwide. If the seizure goes unrecognized, it may lead to fatal accidents. It is important to predict seizure and initiate preventive interventions to reduce the risks. It can be used to alert the patient, relatives, clinicia...

Full description

Saved in:

Bibliographic Details
Published in:	2023 14th International Conference on Computing Communication and Networking Technologies (ICCCNT) pp. 1 - 7
Main Authors:	Nair, Simran, Kumar, C. Santhosh, Muralidharan, Pooja, Gopinath, Siby, Kumar, A. Anand, Parasuram, Harilal
Format:	Conference Proceeding
Language:	English
Published:	IEEE 06-07-2023
Subjects:	Alarm systems Belts CNN Convolutional neural networks Decision fusion DNN ECG Electrocardiography Electroencephalography EMD Epileptic seizure Receivers Sensitivity SVM VMD
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	Epileptic seizure is a neurological disorder affecting 50 million people worldwide. If the seizure goes unrecognized, it may lead to fatal accidents. It is important to predict seizure and initiate preventive interventions to reduce the risks. It can be used to alert the patient, relatives, clinicians and family members can record a video that will guide the neurologist to refine the treatment. Electroencephalogram (EEG) is a standard method for diagnosis. But it is less suitable in an ambulatory setting. A more convenient alternative is to wear an electrocardiogram (ECG) belt around the chest.We explore the possibility of designing an ECG-based patient warning system. We use a deep neural network (DNN) as a baseline system with 65.33%, 64.47%, 67.75%, and 0.6752 of accuracy, sensitivity, specificity, and area under the receiver operating characteristics (AUROC) respectively. We improved the performance with convolutional neural network (CNN) and signal decomposition methods using empirical mode decomposition (EMD) and variational mode decomposition (VMD). We empirically obtained the decision weights for the bottleneck features derived from the best-performing EMD and VMD modes. In the decision fusion, the accuracy, sensitivity, specificity, and AUROC is 84.68%, 69.60%, 93.63%, and 0.8162 respectively which showed an improvement of 19.35%, 5.13%, 25.88%, and 0.141 respectively from the baseline system.
ISSN:	2473-7674
DOI:	10.1109/ICCCNT56998.2023.10306997