A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Epileptic Seizure Detection using Spectral Transformation and Convolutional Neural Networks
https://orcid.org/http://orcid.org/0000-0001-5915-4125
Automatic seizure detection and classification of seizures, as well as identification of pre-ictal activity in the electroencephalogram (EEG), are extremely important in clinical research. This decreases the time it takes to identify seizures and, as a result, improves seizure activity prediction. We propose a computer-aided method to detect the pre-ictal and ictal activity from a multichannel EEG signal. Three pre-processing techniques that are applied to EEG time-domain signals to generate an image database are proposed here. This image database is given as input to the machine learning algorithm for classification. Conversion of a time domain EEG signal to an image is accomplished by extracting EEG signal features such as correlation coefficient, short-time Fourier transform (spectrogram), and mutual information. The processed EEG waveform, which is represented as images, is used to train a convolutional neural network (CNN). The CNN classifies input signals into three classes—Seizure, Normal and Pre-ictal. We used the transfer learning method, which uses Alexnet, a pre-trained CNN architecture, for image training and classification. After training on the Spectrogram, Mutual Information, and Correlation coefficient image representations of the EEG signals, we have obtained a validation accuracy of 99.33%, 95.33%, and 97.5%, respectively.
Epileptic Seizure Detection using Spectral Transformation and Convolutional Neural Networks
https://orcid.org/http://orcid.org/0000-0001-5915-4125
Automatic seizure detection and classification of seizures, as well as identification of pre-ictal activity in the electroencephalogram (EEG), are extremely important in clinical research. This decreases the time it takes to identify seizures and, as a result, improves seizure activity prediction. We propose a computer-aided method to detect the pre-ictal and ictal activity from a multichannel EEG signal. Three pre-processing techniques that are applied to EEG time-domain signals to generate an image database are proposed here. This image database is given as input to the machine learning algorithm for classification. Conversion of a time domain EEG signal to an image is accomplished by extracting EEG signal features such as correlation coefficient, short-time Fourier transform (spectrogram), and mutual information. The processed EEG waveform, which is represented as images, is used to train a convolutional neural network (CNN). The CNN classifies input signals into three classes—Seizure, Normal and Pre-ictal. We used the transfer learning method, which uses Alexnet, a pre-trained CNN architecture, for image training and classification. After training on the Spectrogram, Mutual Information, and Correlation coefficient image representations of the EEG signals, we have obtained a validation accuracy of 99.33%, 95.33%, and 97.5%, respectively.
Epileptic Seizure Detection using Spectral Transformation and Convolutional Neural Networks
https://orcid.org/http://orcid.org/0000-0001-5915-4125
Automatic seizure detection and classification of seizures, as well as identification of pre-ictal activity in the electroencephalogram (EEG), are extremely important in clinical research. This decreases the time it takes to identify seizures and, as a result, improves seizure activity prediction. We propose a computer-aided method to detect the pre-ictal and ictal activity from a multichannel EEG signal. Three pre-processing techniques that are applied to EEG time-domain signals to generate an image database are proposed here. This image database is given as input to the machine learning algorithm for classification. Conversion of a time domain EEG signal to an image is accomplished by extracting EEG signal features such as correlation coefficient, short-time Fourier transform (spectrogram), and mutual information. The processed EEG waveform, which is represented as images, is used to train a convolutional neural network (CNN). The CNN classifies input signals into three classes—Seizure, Normal and Pre-ictal. We used the transfer learning method, which uses Alexnet, a pre-trained CNN architecture, for image training and classification. After training on the Spectrogram, Mutual Information, and Correlation coefficient image representations of the EEG signals, we have obtained a validation accuracy of 99.33%, 95.33%, and 97.5%, respectively.
Epileptic Seizure Detection using Spectral Transformation and Convolutional Neural Networks
J. Inst. Eng. India Ser. B
Cleatus, T. Saneesh (author) / Thungamani, M. (author)
Journal of The Institution of Engineers (India): Series B ; 103 ; 1115-1125
2022-08-01
11 pages
Article (Journal)
Electronic Resource
English
A Convolutional Neural Network for Seizure Detection
| IEEE | 2020
Epileptic Seizure Prediction Methods Using Machine Learning and Deep Learning Models
| Springer Verlag | 2023
Structural crack detection using deep convolutional neural networks
| Elsevier | 2021
Convolutional neural networks based droplet detection method
| Wiley | 2022