Detection of nocturnal convulsive seizures in children by integrating video and accelerometer recordings
Cuppens K., Van de Vel A., Karsmakers P., Bonroy B., Milosevic M., Luca S., Croonenborghs T., Tuytelaars T., Lagae L., Ceulemans B., Van Huffel S., Vanrumste B., ''Detection of nocturnal convulsive seizures in children by integrating video and accelerometer recordings'', 4th Dutc...
Main Authors: | , , , , , , , , , , , |
---|---|
Format: | Conference Object |
Language: | English |
Published: |
2013
|
Subjects: | |
Online Access: | https://lirias.kuleuven.be/handle/123456789/392570 http://www.bme2013.nl/content/company https://lirias.kuleuven.be/bitstream/123456789/392570/1//BME2013_IntegratingAccVid_KrisCuppens.pdf |
id |
ftunivleuven:oai:lirias.kuleuven.be:123456789/392570 |
---|---|
record_format |
openpolar |
institution |
Open Polar |
collection |
KU Leuven: Lirias |
op_collection_id |
ftunivleuven |
language |
English |
topic |
PSI_VISICS |
spellingShingle |
PSI_VISICS Cuppens, Kris Van de Vel, Anouk Karsmakers, Peter Bonroy, Bert Milosevic, Milica Luca, Stijn Croonenborghs, Tom Tuytelaars, Tinne Lagae, Lieven Ceulemans, Berten Van Huffel, Sabine Vanrumste, Bart Detection of nocturnal convulsive seizures in children by integrating video and accelerometer recordings |
topic_facet |
PSI_VISICS |
description |
Cuppens K., Van de Vel A., Karsmakers P., Bonroy B., Milosevic M., Luca S., Croonenborghs T., Tuytelaars T., Lagae L., Ceulemans B., Van Huffel S., Vanrumste B., ''Detection of nocturnal convulsive seizures in children by integrating video and accelerometer recordings'', 4th Dutch bio-medical engineering conference, January 24-25, 2013, Egmond aan Zee, The Netherlands. We focus on the detection of nocturnal convulsive seizures, more specific on hypermotor seizures. This type of seizures are marked by involuntary movement of the patient which can last for multiple seconds. Due to the violent movement the patient can injure himself. The early treatment may prevent consequences such as severe cerebral damage or mortality [1]. Epileptic seizure detection is traditionally done using video/electroencephalogram (EEG) monitoring, which is not applicable in a home situation. In recent years, attempts have been made to detect the seizures using other modalities. In this research we investigate if a combined usage of accelerometers attached to the limbs and video data would increase the performance compared to a single modality approach. Therefore, we used two existing approaches for seizure detection in accelerometers and video. The video detection method makes use of STIP features proposed by Laptev [2], the accelerometer approach makes use of different features from the time and frequency domain. In both approaches, an SVM classifier is used for classification. We combined both approaches (accelero and video) using a linear discriminant analysis (LDA) classifier. The seizure data is acquired at the Pulderbos rehabilitation center for children and youth where epileptic children are monitored and treated. The data is labeled based on the EEG and video. To be able to objectively compare both approaches, based on acceleration and on video, we use the same segmentation of the data for both modalities. The training and testing is done in a 10-fold randomization. This means that we randomly select a number of normal and epileptic movements for the training and test set and perform the modeling and validation 10 times using a different combination of movements, and average out the obtained results. This makes the results less dependent on the division of the data in a training and test set. The same randomizations are used in both approaches and for the acceleration/video integration. In a first test we combined the normalized features from both approaches in an early integration. In a second test, we combined the outputs of both individual classifiers from the video and accelerometer detection in a late integration using an LDA classifier. The output values of both classifiers give a probability of the sequence belonging to the epileptic seizure class, as we use the libsvm implementation for extending SVM to give probability estimates based on Wu et al. [3] and Lin and Weng [4]. The combined detection using the early integration seems to give a lower performance (sensitivity: 83.33%, positive predictive value (PPV): 96.00%) than the accelerometer detection alone (sens: 83.33%, PPV: 100.00%). This means that the video features do not have any added value in this integration. The late integration of both modalities has a small positive influence on the performance. Although there is a decline in the PPV from 100.00% to 97.50% compared with the accelerometer detection, the sensitivity increases to 86.67%. status: published |
format |
Conference Object |
author |
Cuppens, Kris Van de Vel, Anouk Karsmakers, Peter Bonroy, Bert Milosevic, Milica Luca, Stijn Croonenborghs, Tom Tuytelaars, Tinne Lagae, Lieven Ceulemans, Berten Van Huffel, Sabine Vanrumste, Bart |
author_facet |
Cuppens, Kris Van de Vel, Anouk Karsmakers, Peter Bonroy, Bert Milosevic, Milica Luca, Stijn Croonenborghs, Tom Tuytelaars, Tinne Lagae, Lieven Ceulemans, Berten Van Huffel, Sabine Vanrumste, Bart |
author_sort |
Cuppens, Kris |
title |
Detection of nocturnal convulsive seizures in children by integrating video and accelerometer recordings |
title_short |
Detection of nocturnal convulsive seizures in children by integrating video and accelerometer recordings |
title_full |
Detection of nocturnal convulsive seizures in children by integrating video and accelerometer recordings |
title_fullStr |
Detection of nocturnal convulsive seizures in children by integrating video and accelerometer recordings |
title_full_unstemmed |
Detection of nocturnal convulsive seizures in children by integrating video and accelerometer recordings |
title_sort |
detection of nocturnal convulsive seizures in children by integrating video and accelerometer recordings |
publishDate |
2013 |
url |
https://lirias.kuleuven.be/handle/123456789/392570 http://www.bme2013.nl/content/company https://lirias.kuleuven.be/bitstream/123456789/392570/1//BME2013_IntegratingAccVid_KrisCuppens.pdf |
genre |
laptev |
genre_facet |
laptev |
op_relation |
Dutch Conference on Bio-Medical Engineering edition:4 location:Egmond aan Zee, The Netherlands date:24-25 January 2013 https://lirias.kuleuven.be/handle/123456789/392570 PSI_3797 http://www.bme2013.nl/content/company https://lirias.kuleuven.be/bitstream/123456789/392570/1//BME2013_IntegratingAccVid_KrisCuppens.pdf |
op_rights |
219922;public |
_version_ |
1766062684332097536 |
spelling |
ftunivleuven:oai:lirias.kuleuven.be:123456789/392570 2023-05-15T17:07:19+02:00 Detection of nocturnal convulsive seizures in children by integrating video and accelerometer recordings Cuppens, Kris Van de Vel, Anouk Karsmakers, Peter Bonroy, Bert Milosevic, Milica Luca, Stijn Croonenborghs, Tom Tuytelaars, Tinne Lagae, Lieven Ceulemans, Berten Van Huffel, Sabine Vanrumste, Bart 2013-01-24 87451 bytes application/pdf https://lirias.kuleuven.be/handle/123456789/392570 http://www.bme2013.nl/content/company https://lirias.kuleuven.be/bitstream/123456789/392570/1//BME2013_IntegratingAccVid_KrisCuppens.pdf en eng Dutch Conference on Bio-Medical Engineering edition:4 location:Egmond aan Zee, The Netherlands date:24-25 January 2013 https://lirias.kuleuven.be/handle/123456789/392570 PSI_3797 http://www.bme2013.nl/content/company https://lirias.kuleuven.be/bitstream/123456789/392570/1//BME2013_IntegratingAccVid_KrisCuppens.pdf 219922;public PSI_VISICS Conference paper/abstract IMa conference_paper 219922;Conference paper/abstract 2013 ftunivleuven 2014-11-26T12:54:36Z Cuppens K., Van de Vel A., Karsmakers P., Bonroy B., Milosevic M., Luca S., Croonenborghs T., Tuytelaars T., Lagae L., Ceulemans B., Van Huffel S., Vanrumste B., ''Detection of nocturnal convulsive seizures in children by integrating video and accelerometer recordings'', 4th Dutch bio-medical engineering conference, January 24-25, 2013, Egmond aan Zee, The Netherlands. We focus on the detection of nocturnal convulsive seizures, more specific on hypermotor seizures. This type of seizures are marked by involuntary movement of the patient which can last for multiple seconds. Due to the violent movement the patient can injure himself. The early treatment may prevent consequences such as severe cerebral damage or mortality [1]. Epileptic seizure detection is traditionally done using video/electroencephalogram (EEG) monitoring, which is not applicable in a home situation. In recent years, attempts have been made to detect the seizures using other modalities. In this research we investigate if a combined usage of accelerometers attached to the limbs and video data would increase the performance compared to a single modality approach. Therefore, we used two existing approaches for seizure detection in accelerometers and video. The video detection method makes use of STIP features proposed by Laptev [2], the accelerometer approach makes use of different features from the time and frequency domain. In both approaches, an SVM classifier is used for classification. We combined both approaches (accelero and video) using a linear discriminant analysis (LDA) classifier. The seizure data is acquired at the Pulderbos rehabilitation center for children and youth where epileptic children are monitored and treated. The data is labeled based on the EEG and video. To be able to objectively compare both approaches, based on acceleration and on video, we use the same segmentation of the data for both modalities. The training and testing is done in a 10-fold randomization. This means that we randomly select a number of normal and epileptic movements for the training and test set and perform the modeling and validation 10 times using a different combination of movements, and average out the obtained results. This makes the results less dependent on the division of the data in a training and test set. The same randomizations are used in both approaches and for the acceleration/video integration. In a first test we combined the normalized features from both approaches in an early integration. In a second test, we combined the outputs of both individual classifiers from the video and accelerometer detection in a late integration using an LDA classifier. The output values of both classifiers give a probability of the sequence belonging to the epileptic seizure class, as we use the libsvm implementation for extending SVM to give probability estimates based on Wu et al. [3] and Lin and Weng [4]. The combined detection using the early integration seems to give a lower performance (sensitivity: 83.33%, positive predictive value (PPV): 96.00%) than the accelerometer detection alone (sens: 83.33%, PPV: 100.00%). This means that the video features do not have any added value in this integration. The late integration of both modalities has a small positive influence on the performance. Although there is a decline in the PPV from 100.00% to 97.50% compared with the accelerometer detection, the sensitivity increases to 86.67%. status: published Conference Object laptev KU Leuven: Lirias |