Home // COGNITIVE 2023, The Fifteenth International Conference on Advanced Cognitive Technologies and Applications // View article

ECG-based Seizure Prediction Utilizing Transfer Learning with CNN

Authors:
Chia-Yen Yang
Pin-Chen Chen

Keywords: electrocardiography (ECG); Convolutional Neural Network (CNN); seizure prediction; transfer learning

Abstract:
Clinically, electroencephalography (EEG) is the most common tool used to diagnose epilepsy. However, if considering practicality and convenience, electrocardiogram (ECG) is more suitable for use in non-medical institutions. Its problem that needs to be overcome is the improvement of accuracy. Therefore, this study attempted to apply transfer learning strategy to develop a seizure prediction system based on ECG for detecting interictal and preictal periods. We trained a nonpatient specific epilepsy prediction model based on Convolutional Neural Network (CNN), and then used transfer learning to fine-tune parameters with the goal of reducing the model development time and improving the performance for each specific patient. ECG data were obtained from two open-source datasets, the Siena Scalp EEG database and Zenodo, including 13 and 14 patients, respectively. The results show that the patient-specific model with six frozen layers achieved accuracy, sensitivity, and specificity of 100% for nine patients and required only 40 s of training time. By applying transfer learning, the model could directly use raw ECG signals, eliminating the time and manpower in extraction of features and greatly speeding up the training process. Furthermore, it achieved the purpose of personalized and accurate detection that could increase the practicality of seizure prediction in daily life.

Pages: 16 to 20

Copyright: Copyright (c) IARIA, 2023

Publication date: June 26, 2023

Published in: conference

ISSN: 2308-4197

ISBN: 978-1-68558-046-9

Location: Nice, France

Dates: from June 26, 2023 to June 30, 2023