Accelerating the Diagnosis of Epilepsy With Computer Modeling
Kathryn Davis, MD, MTR, director of the Epilepsy Monitoring Unit and assistant professor of Neurology, explained how a startup is using the brain waves of large numbers of people to assess an individual’s risk of seizure disorders more quickly and accurately. She emphasized that those tests aren’t definitive — the scans don’t always show anything unusual in people who do have epilepsy.
Now, a startup is using a database of EEG records from people with and without epilepsy, to create a computer model that builds a network of interactions between the signals produced by the electrodes placed on the head to measure brain activity. The software uses the model to generate a risk score that can help a clinician to assess an individual’s risk of epilepsy.
The symptoms that many people associate with someone having an epileptic seizure — losing consciousness, biting the tongue or even briefly ceasing to breathe — would make it seem that diagnosing this condition would be a simple matter.
But diagnosis is a tricky process that usually takes months — and might never be definitive. People with epilepsy describe their diagnostic journey as “long, confusing, unclear and uncertain”, says John Terry, a mathematician at the University of Birmingham, UK
Terry is co-founder and managing director of Neuronostics in Bristol, UK, a start-up that aims to shorten this journey. The company’s patented software, BioEP, analyses electroencephalography (EEG) recordings of brain activity that are routinely collected during the first appointment with a neurologist. It can identify epilepsy from a single recording, with a similar sensitivity to that of clinicians who conduct numerous tests over many months, says Terry.
The implications are enormous. More than 65 million people, nearly 1% of the global population, have epilepsy, and the condition has serious associated risks. Quicker and more reliable diagnosis could lead to earlier treatment and could substantially reduce the potentially severe impact on people’s quality of life.
“This technology could be a game changer for epilepsy diagnosis,” says Exeter-based Simon Privett, an epilepsy-awareness advocate at the UK charity Epilepsy Action, who has worked closely with Terry and his team.
Beyond the standard scans
Epileptic seizures are driven by unusual electrical activity in the brain. A diagnosis is made on the basis of clinical history, and EEG and magnetic resonance imaging (MRI) scans help to confirm the diagnosis, says Kathryn Davis, a neurologist specializing in epilepsy at the University of Pennsylvania in Philadelphia.
But Davis emphasizes that those tests aren’t definitive — the scans don’t always show anything unusual in people who do have epilepsy.
Among people who have had a seizure but whose scans don’t show unusual activity, the risk of a second seizure is only around 30%, Davis says. She usually doesn’t prescribe anti-epileptic drugs at that point. “I’m just counselling them to wait and see if you have another potentially life-threatening event,” she says. “And if you do, then we’ll treat you. That’s an extremely common scenario.”
Currently, clinicians examine EEG readouts and look for precursors of seizures such as spikes in activity or brief bursts of unusual pathological activity. They also look for seizures. “The trouble is that these are all quite rare events,” Terry says. So clinicians usually begin a process to eliminate psychological or physical causes of the seizures — monitoring people for progressively longer periods of time, and upping stresses that can trigger seizures, such as lack of sleep, to pick up signals of epilepsy.
Terry and his collaborators at Neuronostics instead use 20 seconds of seizure-free data from a routine EEG scan. Using a database of EEG records from people with and without epilepsy, the researchers have created a computer model that has built a network of interactions between the signals produced by the electrodes placed on the head to measure brain activity, called EEG channels1 (see ‘Simulating seizure risk’). This model, Terry explains, describes “transitions between background states in the brain that are apparently healthy, and seizure-like states that are apparently pathological”. The Neuronostics software uses the model to generate a risk score that can help a clinician to assess an individual’s risk of epilepsy.
Simulating seizure risk
With a conventional evaluation of an initial EEG scan, a clinician can correctly diagnose roughly 25% of people with epilepsy, according to Terry. Unpublished research with the Neuronostics model shows that it performs much better, detecting epilepsy in 60% of those who do have the condition, and making the right assessment in 87% of cases.
Accelerating the diagnosis of epilepsy has two major advantages. First, it avoids the need for long-term testing, and the associated disruption for those being examined. Second, it reduces the number of seizures a person is likely to have — delays in diagnosis and treatment mean that seizures will keep happening, and, generally, the more seizures someone has, the more likely they are to have more.
If fully validated, the ability of Neuronostics’ method to outperform clinicians’ interpretations of initial EEGs could prove very helpful for diagnosis, says Davis.
Delivering on diagnostics
Neuronostics’ system has been long in gestation. Terry and his colleagues first reported2 the method in 2012. Terry then discussed the research with a commercial technology specialist at the University of Exeter (where Terry was then based) who was impressed with the technology and its potential impact. The university applied for patent protection and, in 2018, aided the company’s formation. It also helped to secure a series of UK government grants — most recently, two awards totalling £870,000 (US$1.2 million) in September 2020 from the National Institute for Health Research.
In March, the company raised £300,000 in a seed round of financing. It now has four full-time employees and is hiring, among other roles, a head of marketing.
Neuronostics estimates there is a potential market of 3.5 million EEGs annually in the United States and Europe, and it plans to charge a per-use fee of $400 in the United States and the equivalent of $280 in the United Kingdom (the fee is lower in the United Kingdom because of requirements under its government sponsorship). The company received regulatory approval for BioEP in Europe and the United Kingdom in May, with expansion to the United States planned for late 2022.
The diagnosis of epilepsy might not, however, prove to be the software’s most valuable offering. The start-up also sees potential for determining prognosis for people taking epilepsy treatments, says mathematician Wessel Woldman at the University of Birmingham, who is co-founder and scientific director at Neuronostics. Indeed, although some neurologists remain sceptical about EEG analysis for diagnosis, “when it comes to prognosis, there’s a lot more of a realization that computer-assisted methods could really contribute something meaningful”, Woldman says.
About one-third of people diagnosed with epilepsy get no benefit from any available drug. For the remaining two-thirds, finding the right treatment can take months. Neuronostics analyses might allow clinicians to observe the effects of a medication much more quickly, which could cut down on seizures and minimize the use of medications with significant side effects, Woldman says. He is collaborating with clinicians at St Vincent’s Hospital in Fitzroy, Australia, to test the approach. Very early results are encouraging, he says.
Neuronostics is also developing a smartphone app called ConnectEP that works with a new generation of inexpensive, gel-free wireless EEG headsets designed for home use. The resulting data won’t be clinical-grade but could still prove highly useful to neurologists, Terry says. The company hopes to release the app by the end of 2021.
Later generations of ConnectEP could also produce useful short-term forecasts of risk of seizures — an exciting possibility, says Privett, who is involved in testing prototypes. Overall, Privett says, the potential impact of Neuronostics technology for people with epilepsy is huge.
As it commercializes its technology, Neuronostics is maintaining its tight focus on feedback from all the stakeholders in epilepsy treatment — especially those with the condition. “Our ethos”, says Terry, “is to ensure that our models and experiments and clinical work develop very closely together.”