Eventos Anais de eventos
ENEBI 2022
VII Encontro Nacional de Engenharia Biomecânica
Epilepsy Seizure Detection Using Time and Frequency Domain EEG Signal via Convolutional Neural Network
Submission Author:
Cristina Natalia Espinosa Martinez , MG , Brazil
Co-Authors:
Cristina Natalia Espinosa Martinez, Clarissa Lima Loures, Rudolf Huebner
Presenter: Cristina Natalia Espinosa Martinez
doi://10.26678/ABCM.ENEBI2022.EEB22-0019
Abstract
This work aims to develop an epilepsy detection system based on Convolutional Neural Networks (CNNs). The CNN model consists of 8-layers and receives three input signal types. Moreover, to evaluate the model performance is applied two training cases. In case 1, it was created one training set per patient. In case 2, one training set was built containing different seizure types. Hence, in case 1, the recall and precision achieved with both inputs in the time and frequency domain were: 44.1% / 43.8%, and 44.4% / 45.1%, respectively; whereas with the input in the time-frequency domain, the recall and precision were 81.6% and 92.5%, respectively. On the other hand, in case 2, evaluating the test session of the same patients used to train the model, the recall was 82% and precision was 86%. Demonstrating that the CNN model learns each person's brain-behavior, generalizing the knowledge, and detecting any seizure type.
Keywords
epilepsy, Seizure detection, Convolutional neural networks, short time fourier transform, Electroencephalogram
DOWNLOAD PDF VIEW PRESENTATION
