EpiScalp Cuts Epilepsy Misdiagnoses From 54% to 17% in Normal EEGs

A computational tool called EpiScalp ruled out 96% of false positive epilepsy diagnoses when Johns Hopkins researchers reanalyzed routine scalp EEGs (electroencephalograms, recordings of brain electrical activity via electrodes on the scalp) that had supported an initial epilepsy call. Potential misdiagnoses in the study cohort fell from 54% to 17%. The work appears in Annals of Neurology, published January 22, 2025.

The retrospective analysis covered 198 patients drawn from five medical centers: Johns Hopkins Hospital, Johns Hopkins Bayview Medical Center, University of Pittsburgh Medical Center, University of Maryland Medical Center, and Thomas Jefferson University Hospital. Of these, 91 had epilepsy; the remaining 107 had non-epileptic conditions that mimic it.

"Even when EEGs appear completely normal, our tool provides insights that make them actionable," said Sridevi V. Sarma, professor of biomedical engineering at Johns Hopkins, who led the work. "We can get to the right diagnosis three times faster because patients often need multiple EEGs before abnormalities are detected, even if they have epilepsy."

Why routine EEGs miss epilepsy

Standard initial epilepsy evaluation relies on scalp EEG during a typical 20-to-40-minute recording. Seizures rarely occur within that window, and the captured signals are noisy. Interpretation is therefore subjective and error-prone even for specialists. False positives occur in roughly 30% of cases globally, according to the Johns Hopkins summary, exposing patients without epilepsy to anti-seizure medication side effects, driving restrictions, and other burdens.

Sarma's team took a different route: instead of hunting for epileptiform discharges, they studied what the brain does between seizures. Their hypothesis holds that some brain regions act as natural inhibitors of seizure activity. "It's like the brain's immune response to the disease," Sarma said.

Earlier work by the same group, using intracranial EEG (electrodes placed inside the skull), demonstrated that neighboring regions suppress the seizure onset zone during interictal periods. EpiScalp translates that finding to routine scalp recordings.

How the algorithm reads a normal EEG

EpiScalp applies dynamic network models — algorithms that map how brain regions interact rather than how individual electrodes behave — to a single routine EEG. First author Patrick Myers, a doctoral student in biomedical engineering at Johns Hopkins, described the signature: independent nodes generating excessive activity while a second region suppresses them, with the whole cluster decoupled from the rest of the brain's network.

"A healthy person shouldn't have that," Myers said.

What does this change for the reading room and monitoring unit?

If validated prospectively, EpiScalp could turn a normal 20-to-40-minute scalp EEG from a nondiagnostic study into an actionable one, reducing repeat EEG orders and shortening the diagnostic path. Co-senior author Khalil Husari, assistant professor of neurology at Johns Hopkins, framed the clinical stakes directly: "These patients experienced side effects of the anti-seizure medication without any benefit because they didn't have epilepsy. Without the correct diagnosis, we can't find out what's actually causing their symptoms."

Some misdiagnoses stem from EEG over-reading, as clinicians weigh the danger of a second seizure; others involve non-epileptic seizures, which often respond to therapies other than epilepsy medication. EpiScalp targets both failure modes by extracting network-level markers invisible to conventional visual review.

EpiScalp remains investigational — it is not FDA-cleared — and the current results come from retrospective reanalysis, not a prospective trial. A larger prospective validation study is now underway across three epilepsy centers. The team filed a patent for the technology in 2023.