A weather station for epilepsy

In this video abstract, the authors review the discovery of cycles of epileptic

In this video abstract, the authors review the discovery of cycles of epileptic brain activity. They explain the computational advances that made seizure forecasting possible and the practical implications of providing personalised seizure risk stratification.

To do this, Baud collaborated with Vikram Rao, neurologist at UCSF, to obtain neuronal activity data collected over several years using devices implanted long-term in the brains of patients with epilepsy. After confirming that there were indeed cycles of cerebral epileptic activity, the scientists turned their attention to statistical analysis. This approach helped highlight a phenomenon known as the pro-ictal state where the probability of the onset of a seizure is high.

"As with weather disturbances, there are several time scales in epileptic brain activity", points out Dr Baud. "The weather is influenced by the cycle of the seasons or day and night. On an intermediate scale, when a weather front approaches, the probability that it will rain increases for several days and is, therefore, more predictable. These three scales of cyclic regulation also exist for epilepsy."

The right timeframe

The electrical activity in the brain is a reflection of the cellular activity of its neurons, more precisely their action potentials, electrical signals propagating along the neural network to transmit information. Ac­tion potentials are well known to neuroscientists, and their proba­bility can be modelled using mathematical laws.

"We adapted these mathematical models to the epileptic discharges to find out whether they heralded or inhibited a seizure", explains Dr Proix.

To boost the predictive reliability, recordings of brain activity over very long periods were required. Using this approach, fronts with a high probability of seizure lasting several days could be determined for a majority of pa­tients, making it possible to predict seizures several days in advance in some. With brain activity data collected over periods of at least six months, seizure prediction is informative for two-thirds of patients.

The analytical approach is sufficiently light to allow the transmis­sion of data in real time on a server or directly on a microprocessor with a device small enough to be implanted in the skull.

The paper ’ Forecasting seizure risk in adults with focal epilepsy: a development and validation study ’ by Timothée Proix, Wilson Truccolo, Marc G Leguia, Thomas K Tcheng, David King-Stephens, Vikram R Rao, Maxime O Baud is published in The Lancet Neurology.