US researchers develop new algorithm to predict seizures
Being able to accurately predict when an epilepsy patient is about to have a seizure could be one step closer to reality thanks to a new software algorithm, according to US researchers.
In the past, technology that interprets electroencephalogram (EEG) readings to work out what happens in the brain at the onset of a seizure has been demonstrated with some success. However, it tends to require impractically large volumes of pre-recorded EEG data to be effective.
Conversely, the new algorithm – developed by Shouyi Wang of the University of Texas, Wanpracha Art Chaovalitwongse of the University of Washington and Stephen Wong of the University of Medicine and Dentistry of New Jersey – uses an ‘adaptive learning’ technique, refining the process of prediction as more data becomes available.
While method this might still require long-term EEG recordings to be truly useful, the researchers theorise that a portable device with discrete electrodes might one day be introduced to address this – perhaps worn “under a cap or hat”.
This data could then be fed directly into the algorithm and used to fine-tune the prediction model for that particular patient.
“Our experimental results showed that the adaptive prediction scheme could achieve a consistent better prediction performance than a chance model and the non-updating system,” the team commented.
“This study confirmed that the concept of using adaptive learning algorithms to improve the adaptability of seizure prediction is conceivable… If a seizure-warning device is ever to become a reality, adaptive learning techniques will play an important role.”
While many people with epilepsy use drugs to keep their symptoms in check, as many as a third of all patients find their seizures do not respond to these therapies.
An device that could accurately predict when they are about to have a seizure could make it possible for these individuals to live normal lives, taking part in activities like driving and operating machinery that might be hazardous otherwise.
Posted by Steve Long