An ongoing clinical trial aims to test whether digital models built using brain-scan data can help to identify where seizures originate.
Virtual models representing the brains of people with epilepsy could help to enable more-effective treatments of the disease by showing neurosurgeons precisely which zones are responsible for seizures.
The models, created using a computational system known as the Virtual Epileptic Patient (VEP), have been developed as part of the Human Brain Project (HBP), a ten-year European initiative focused on digital brain research. The approach is being tested in a clinical trial called EPINOV, to evaluate whether it improves the success rate of epilepsy surgery.
“It’s an example of personalized medicine,” says Aswin Chari, a neurosurgeon at University College London. VEP uses “the patient’s own brain scans [and] the patient’s own brainwave-recording data to build a model and improve our understanding of where their seizures are coming from”.
Epileptic seizures are brought on by abnormal brain activity, and around one-third of the 50 million people living with epilepsy worldwide do not respond to anti-seizure drugs.
“For those people, surgery is a huge game changer,” says Chari. It aims to free patients from seizures by removing parts of the epileptogenic zone — the brain region that is thought to initiate seizures.
To identify the epileptogenic zone, clinicians currently use scanning techniques such as magnetic resonance imaging (MRI) and electroencephalogram (EEG) to investigate brain activity. They also perform stereoelectroencephalography (SEEG), which involves placing up to 16 electrodes, each 7 centimeters long, through the skull to monitor the activity of specific areas for 1–2 weeks.
But SEEG captures only high-frequency currents in the brain. It doesn’t detect lower-frequency activity, which occurs in 20% of seizures. “A lot of people with epilepsy don’t have visible problems on the scan,” says Linda Douw, a neuroscientist at Amsterdam University Medical Centre in the Netherlands.
This makes the precise localization of the epileptogenic zone a considerable challenge, and affects the outcome of the surgery: the success rate at preventing seizures is currently only 50%. “The failure of the surgery is often attributed to a misidentification of the epileptogenic zone,” says Viktor Jirsa, a neuroscientist at Aix-Marseille University, France, who spoke about the project at a summit on the HBP in Marseille last week.
Jirsa and his colleagues hope that artificial intelligence (AI) can offer a more accurate way to identify the epileptogenic zone. The technique they developed involves creating a ‘digital twin’ of a person’s brain by feeding a virtual network based on the human brain with an individual’s MRI, EEG and SEEG data obtained during the presurgical routine.
The researchers then use AI-based simulations on the model to mimic brain activity and determine the zone responsible for seizures. They can also simulate the effects of performing a particular surgery and use those simulations to determine precisely which brain regions to remove, so as to stop a person’s seizures while minimizing the risk of damage. They described this approach in a January paper in Science Translational Medicine.
The current VEP model has a spatial resolution of one square millimeter, but the researchers are working to increase it by a factor of 1,000. “The consequence is that it takes 1,000 times longer,” says Jirsa, and this poses computational challenges in terms of accelerating the processing codes. “We have taken full advantage of the high-performance computing infrastructure of the HBP,” Jirsa adds.
The EPINOV trial began in June 2019, and so far, 356 people across 11 French hospitals have enrolled. The researchers will follow up with study participants for one year after their surgery and evaluate whether VEP improves surgery outcomes, says Jirsa.
“The outcome after a year is a pretty good indicator for how things are going to be in the longer run,” says John Duncan, a neurologist at the UCL Queen Square Institute of Neurology, UK. But “you have to expect that some of those people who are seizure-free for a year will not still be seizure-free at five years”, he warns.
Although the digital-twin approach holds promise, there are limitations. The pattern of seizures can change over time, and they can spread along atypical pathways. They can “come from regions that are not sampled by the SEEG”, says Douw, which means that they cannot be modelled in VEP. “There is a sort of bias in the way that the brain was sampled, [which] might decrease the value of the model in clinical practice.”
The VEP model might also recommend performing surgery on a larger area of the brain than would normally be operated on, says Duncan, and it is going to require strong evidence to convince clinicians that such risks are worth taking. “If it’s an eloquent area of the brain, say in the left temporal lobe in somebody who is left language dominant, that might carry a higher risk of impairing some kind of eloquent function like word finding or verbal memory,” he says.
“The most difficult thing is proving that these algorithms or these new models are actually clinically beneficial,” says Chari.
Enrolment in the EPINOV trial will finish this year, and analysis of the data will begin in late 2024, once the last participants complete their one-year follow-up.
The researchers hope that if it shows promise for people with epilepsy, the virtual-brain approach could be used to study other conditions, such as Parkinson’s disease and multiple sclerosis. “There are lots of diagnostic potential uses in the future of the same mechanism that we use here,” says Jirsa.
Source: nature.com, Miryam Naddaf