Abstract Details

Title Automatic Seizure Detection Using Multi-Resolution Dynamic Mode Decomposition

Topic Epilepsy/Clinical Neurophysiology (EEG)

Presentation(s) S48 - Epilepsy/Clinical Neurophysiology (EEG) III (1:22 PM-1:33 PM)

Poster/Presentation
Number 003

Objective Accurate detection of seizure using a mathematical algorithm.

Background Seizure is one of the most prevalent neurological disorder around the world. Epilepsy is marked by intermittent seizures, the detection of which can be a challenging problem. Therefore, reliably detecting the onset of seizures specially in settings without immediate access to a neurologist has evoked an interest over the last few years. Major leaps in the domain of machine learning and signal processing methods including Dynamic mode decomposition, Mallat's scattering transform, Support vector machines, Artificial neural networks have been used to detect seizure. Our study is the first ever to use multi-resolution Dynamic Mode Decomposition (mrDMD) for detecting seizure.

Design/Methods We used mrDMD, a data-driven dimensionality reduction algorithm. This method can aptly separate a complex system into an order of time-scale components at different resolutions. We have applied this algorithm on two large open source scalp EEG datasets. Necessary post-processing steps including Run length smoothing filter, Aggregation and Consolidation were applied to reduce the false alarm rate while maintaining high sensitivity and specificity. Matlab was used to obtain temporal features and for implementation of algorithm.

Results We analyzed over one thousand hours of combined EEG data from the two large open databases. We achieve a sensitivity of 0.94 and 0.96, specificity of 0.98 and 0.986 and false alarm rate of 0.84 and 0.795 per hour for seizure detection from the two datasets respectively.

Conclusions Our methodology achieves the highest sensitivity without compromising specificity for seizure detection using two large datasets. Our methodology manifested its applicability to efficiently discern seizure and non-seizure from EEG scalp recordings. Moreover, mrDMD also helps in reducing the computational complexity of the system as compared to previously used methods such as DMD. This method can be used to detect seizure accurately and remotely without an on site neurologist.

Authors/Disclosures
Muhammad F. Bilal, MD, FAAN (Indiana University West) PRESENTER	Dr. Bilal has nothing to disclose.
	No disclosure on file
	No disclosure on file

��ɫ��

��ɫ��