Article, originally published on PennToday
By applying tools of machine learning and network analysis, the Davis Lab in the Penn Epilepsy Center was assisted by a team of Penn interns this summer to target the ‘missing electrode problem,’ identifying regions of the brain that cause epilepsy.
One percent of the U.S. population, or three million people, lives with epilepsy. Approximately one-third of those will have a drug-resistant form of epilepsy that, often, demands surgery. The challenge: Localizing where their seizures are coming from before surgery has historically been done with the implantation of EEG electrodes in the brain—an extraordinarily invasive procedure that comes with its own set of limitations.
Enter, the lab of Kathryn Davis, an assistant professor of neurology in the Penn Epilepsy Center at the Perelman School of Medicine.
“The problem with intracranial EEG electrodes is mainly a sampling issue. To solve this problem, we are leveraging information captured from whole-brain neuroimaging to better localize the seizure onset zone,” says Andrew Revell, a fifth-year MD/Ph.D. student in the Davis Lab, who explains that implanted electrodes can miss the areas implicated in seizure generation. “When planning for epilepsy surgery to remove the seizure onset zone, you want to precisely localize the seizure generating areas and avoid eliminating any functional brain tissue, such as areas associated with hand movement or language generation. This is where an MRI can help.”
The Davis lab is studying how an MRI of the brain may help with the sampling issue of intracranial EEG, or the so-called “missing electrode problem,” where surgical limitations and safety restrict implantation of the entire brain. Patients routinely undergo an MRI of their brain before implantations, but they may also participate in research scans to acquire different imaging sequences, such as High-Angular Resolution Diffusion Imaging (HARDI). With these MRI sequences, the lab is modeling how the brain is connected, building networks of the brain, and making predictions of seizure activity in regions where electrodes were not implanted. They take an interdisciplinary approach, drawing from their expertise in imaging analysis, machine learning, network analysis, and signal analysis to solve a very relevant clinical problem.