Classification of electrographic seizures using deep learning approach
Seizures are referred to as a type of neurological disorder which is described as an unexpected, unvoluntary electrical disturbance in neurons of the brain cells resulting in the uncontrolled changes in the behavior, physical movements, feeling and consciousness level of patients. As neurologists cl...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis-Master by Coursework |
| Language: | English |
| Published: |
Nanyang Technological University
2019
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/136499 |
| _version_ | 1811686647692001280 |
|---|---|
| author | Bhowmik, Kaushik |
| author2 | Justin Dauwels |
| author_facet | Justin Dauwels Bhowmik, Kaushik |
| author_sort | Bhowmik, Kaushik |
| collection | NTU |
| description | Seizures are referred to as a type of neurological disorder which is described as an unexpected, unvoluntary electrical disturbance in neurons of the brain cells resulting in the uncontrolled changes in the behavior, physical movements, feeling and consciousness level of patients. As neurologists classify seizures into two types which are electroclinical and EEG only seizures because for each type of seizure, the neurologists prescribe different kinds of medicines to the patients for their recovery and relief. And to make this classification between these two type of seizures, the neurologists have to go through the long EEG recordings of patients and even compare the EEG recording during the seizure event with the video of patient to make this classification which is very time consuming and even sometimes seizure events are of such low intensity that they can easily be missed by neurologists. These reasons motivated me to use the deep learning method to make this classification between these two types of seizures which will save a lot of time of neurologists. In this project, I have used the Temple University Hospital dataset named (NEDC TUH EEG Seizure (v1.3.0)) and in this dataset, I have utilized the data from the folder “train_02”. The EEG recordings were made in the TCP montage (Temporal Central Parasagittal Bipolar Montage). Since it is the preferred way of viewing seizure data at Temple University Hospital. The EEG recordings at Temple University Hospital were recorded at different sampling frequencies 250Hz, 256Hz, 400Hz, 512Hz and for the uniform analysis I have resampled the data at 128Hz. In the (train_02) folder of the dataset, there are 126 seizures of the electroclinical type and 255 seizures of the EEG-only type. And then I have segmented each seizure into segments of 5 seconds duration, which resulted in a total of 88220 segments of the electroclinical type and 295960 segments of the EEG-only type and then after this for the balanced training. I randomly selected 88220 segments of data from 295960 segments. Then the seizures segments of the electroclinical and the EEG-Only type were then fed to the Deep Learning Convolutional Neural Network for the training of the network. The deep learning model consists of four 1-D convolutional layers, four max-pooling layers, two dense layers and 1 flattening layer. The data is then split into training and test data in which 80% of data is used for training and 20% for testing. The model uses 3 fold analysis for better training of the model. The layers use “Relu” and “hard_sigmoid” as the activation function. In this model, I have used the ”Adam” optimizer and the performance of the model is monitored by the “Validation Accuracy”. I have achieved an accuracy of 77.938% with this model and this is the first time that an attempt has been made by using the deep learning technique for doing the classification between electroclinical and EEG-Only seizures, which will save a lot of neurologist’s time which is consumed in doing this classification manually by comparing the EEG recording with the video of patient. |
| first_indexed | 2024-10-01T05:03:45Z |
| format | Thesis-Master by Coursework |
| id | ntu-10356/136499 |
| institution | Nanyang Technological University |
| language | English |
| last_indexed | 2024-10-01T05:03:45Z |
| publishDate | 2019 |
| publisher | Nanyang Technological University |
| record_format | dspace |
| spelling | ntu-10356/1364992023-07-04T15:44:07Z Classification of electrographic seizures using deep learning approach Bhowmik, Kaushik Justin Dauwels School of Electrical and Electronic Engineering JDAUWELS@ntu.edu.sg Engineering::Electrical and electronic engineering Seizures are referred to as a type of neurological disorder which is described as an unexpected, unvoluntary electrical disturbance in neurons of the brain cells resulting in the uncontrolled changes in the behavior, physical movements, feeling and consciousness level of patients. As neurologists classify seizures into two types which are electroclinical and EEG only seizures because for each type of seizure, the neurologists prescribe different kinds of medicines to the patients for their recovery and relief. And to make this classification between these two type of seizures, the neurologists have to go through the long EEG recordings of patients and even compare the EEG recording during the seizure event with the video of patient to make this classification which is very time consuming and even sometimes seizure events are of such low intensity that they can easily be missed by neurologists. These reasons motivated me to use the deep learning method to make this classification between these two types of seizures which will save a lot of time of neurologists. In this project, I have used the Temple University Hospital dataset named (NEDC TUH EEG Seizure (v1.3.0)) and in this dataset, I have utilized the data from the folder “train_02”. The EEG recordings were made in the TCP montage (Temporal Central Parasagittal Bipolar Montage). Since it is the preferred way of viewing seizure data at Temple University Hospital. The EEG recordings at Temple University Hospital were recorded at different sampling frequencies 250Hz, 256Hz, 400Hz, 512Hz and for the uniform analysis I have resampled the data at 128Hz. In the (train_02) folder of the dataset, there are 126 seizures of the electroclinical type and 255 seizures of the EEG-only type. And then I have segmented each seizure into segments of 5 seconds duration, which resulted in a total of 88220 segments of the electroclinical type and 295960 segments of the EEG-only type and then after this for the balanced training. I randomly selected 88220 segments of data from 295960 segments. Then the seizures segments of the electroclinical and the EEG-Only type were then fed to the Deep Learning Convolutional Neural Network for the training of the network. The deep learning model consists of four 1-D convolutional layers, four max-pooling layers, two dense layers and 1 flattening layer. The data is then split into training and test data in which 80% of data is used for training and 20% for testing. The model uses 3 fold analysis for better training of the model. The layers use “Relu” and “hard_sigmoid” as the activation function. In this model, I have used the ”Adam” optimizer and the performance of the model is monitored by the “Validation Accuracy”. I have achieved an accuracy of 77.938% with this model and this is the first time that an attempt has been made by using the deep learning technique for doing the classification between electroclinical and EEG-Only seizures, which will save a lot of neurologist’s time which is consumed in doing this classification manually by comparing the EEG recording with the video of patient. Master of Science (Computer Control and Automation) 2019-12-20T02:54:36Z 2019-12-20T02:54:36Z 2019 Thesis-Master by Coursework https://hdl.handle.net/10356/136499 en application/pdf Nanyang Technological University |
| spellingShingle | Engineering::Electrical and electronic engineering Bhowmik, Kaushik Classification of electrographic seizures using deep learning approach |
| title | Classification of electrographic seizures using deep learning approach |
| title_full | Classification of electrographic seizures using deep learning approach |
| title_fullStr | Classification of electrographic seizures using deep learning approach |
| title_full_unstemmed | Classification of electrographic seizures using deep learning approach |
| title_short | Classification of electrographic seizures using deep learning approach |
| title_sort | classification of electrographic seizures using deep learning approach |
| topic | Engineering::Electrical and electronic engineering |
| url | https://hdl.handle.net/10356/136499 |
| work_keys_str_mv | AT bhowmikkaushik classificationofelectrographicseizuresusingdeeplearningapproach |