Home•Our Research•Publications•Identification of EEG features during status epilepticus for prediction of emergent epilepsy phenotype in the mouse intra-amygdala kainic acid model using supervised learning

Identification of EEG features during status epilepticus for prediction of emergent epilepsy phenotype in the mouse intra-amygdala kainic acid model using supervised learning

Thematic Area:
Therapeutics
Disease Area:
Epilepsy
Collaborators:
Academic
Read full publication

Background

Injury to the brain is known to cause epilepsy. At an individual level, it remains difficult to predict your risk of epilepsy or how frequent seizures will be in the future. Can computer models and artificial intelligence help?

Research

We used recordings of brain signals collected by electroencephalography (EEG) from a mouse model of epilepsy. These we sent through a computer science tool called machine learning to search for EEG signals that could predict the severity of epilepsy. The model performed well, correctly identifying the type of epilepsy (mild, moderate, severe) that would emerge based on only 40 minutes of EEG recorded before the first epileptic seizure. When we tested the predictor using a set of previously unseen EEG recordings, the algorithm was less accurate.

Potential Impact

The findings tell us there are early bio-electrical features that signal the form of epilepsy that will arise in the future. Further adjustments of the model and testing using EEG from other sources are needed to improve accuracy. The results may help with the design of preclinical drug screening studies.

Read full publication