A tool to automatically detect seizures



I would like to introduce everyone to a new program which can detect seizures automatically in a rodent model of acquired and genetic epilepsy. I found about this program during one of the poster sessions. The title of the poster was “A novel user-friendly automated tool to accurately detect seizures in rodent models of acquired and genetic epilepsy”, and it was presented by Armen Sargsyan from the Department of Medicine, The University of Melbourne, Melbourne, Australia.

Video-EEG monitoring in epileptogenesis and chronic epilepsy rodent models is an important tool in pre-clinical drug development of new therapies. Many researchers prefer identifying seizures by themselves, by visually inspecting the EEG and comparing it with the behavior in the video. The authors developed a computation tool which is easy to use and is reliable for detection of electrographic seizures from long EEG recordings. They call the tool Spectral Band Index Tool (SBIT).

SBIT uses an advanced time-frequency analysis detecting the EEG segments with excessive activities in many frequency bands. This is calculated by using Spectral Band Index (SBI), which is a function of time, using time window. The SBI is defined either as the maximum or the average of the window power spectrum within the given frequency band. The authors tested the tool on two acquired and two genetically spontaneous seizures in chronic epilepsy rat models. They found that ictal EEG in these models contains a strong component with frequency of about 20-21 Hz, however, this component was missing in the inter-ictal EEGs.

The SBIT algorithm has two main steps. The first step calculates the SBI and selects the events of interest. In the second step, users visually examine the selected events and classify them as seizures or artefacts. Other additional steps involve artefact removal and parameter control.

The authors analyzed recordings from 99 rats which contained ~ 10,000 seizures, and the program recorded 100% of seizures in all the models. The longest time spent by the user for processing one animal data was about 5 minutes per day of record, and processing of a single day of recording of one animal took about a minute.

This tool reduced the time needed for identification of seizures in long term EEG recordings by ~95%. In conclusion, the authors say that the SBIT program significantly reduces the processing time. The tool is very easy to use as it requires minimal interference of the user. The authors monitored ~ 10,000 seizures and they claim to have not missed a single seizure.

My PhD was based on epilepsy and I regularly monitored epileptic seizures. I know the pain behind monitoring and analysis of the seizures. It takes a lot of time to go through a whole day of EEG recordings, and when you have multiple animals to analyze, you spend most of the day doing the same. This program might be the answer to my question of detecting seizures automatically. However, even if the program determines the seizures, I would still go back and confirm the EEG activity.

Chinmaya Sadangi
Twitter: @addictivebrain
Neuronline: @csadangi