• More than 500,000 French patients are suffering from epilepsy and for 30% of them drugs have no effect in regulating seizures.
  • In some cases, the only solution is surgical with the removal of the area of the brain responsible for the disease.
  • Locating it precisely is a tedious job for specialists. But three Toulouse researchers have developed an algorithm capable of detecting it automatically and quickly.

Epilepsy, a constraining neurological condition, is not a rare disease. It affects about 1% of the world's population and more than 500,000 patients in France, half of whom are under 20 years old. In most cases, neuroleptics can reduce or stop seizures. But "about a third of patients are resistant to drugs and for tens of thousands of them, epilepsy is a real handicap in their daily lives," says Ludovic Gardy, a CNRS engineer at the Brain and Cognition Research Center (CerCo) in Toulouse, whose thesis work is about to lead to the creation of a start-up that could improve the lot of these "drug-resistant" patients.

For the latter and if they meet certain conditions, there is a surgical solution as a last resort: the removal, directly in the brain, of the so-called "epileptogenic area", at the origin of the seizures. But this option, which concerns "nearly a thousand patients per year in France", is laborious, very tedious. Because, often, to precisely locate the area to be operated, it is first necessary to be hospitalized "ten to fifteen days" of days for the implantation of a dozen intracerebral electrodes – long rods of about ten centimeters on average – equipped with sensors to record neuronal activity. The specialist then finds himself with bundles of electroencephalogram (EEG) curves that he must decipher, compile to establish his diagnosis. A "colossal" and tedious investigation that can take several months. "It's a bit like the epileptologist having to swallow a 500-page novel every day, make a summary of it and, in addition, highlight each appearance of the word 'house'," says Ludovic Gardy.

One hospitalization instead of two?

The specialist can use different epilepsy biomarkers as a "compass". According to "the current literature", the most effective are "fast ripples", very rapid pathophysiological oscillations. But they are so microscopic that they are "strictly invisible to the naked eye" on the curves.


And this is where the algorithm developed by Ludovic Gardy and his two thesis supervisors Emmanuel Barbeau of CerCo and Christophe Hurter, a researcher in artificial intelligence at the National School of Civil Aviation (ENAC), comes in. "Not only is the algorithm able to "automatically detect fast ripples, but we wrapped it in software that allows classic data manipulation," says the engineer. This "tool" was tested a posteriori on the EEGs of about thirty "implanted" patients. He located the fast ripples well in the areas determined by the specialists thanks to their classical biomarkers visible to the naked eye. But in days instead of months.

By the summer, the three researchers must create the start-up Avrio MedTech, a subsidiary of the CNRS, which will improve and then commercialize the software. For "implanted" epileptics, EEGs and surgery could be done in a single hospital stay, whereas the process can currently take a year.

  • epilepsy
  • Health
  • Sciences
  • Artificial Intelligence (AI)
  • Toulouse
  • Occitania
  • Midi-Pyrénées
  • Haute-Garonne