Skip Navigation
MUSC mobile menu

Resident’s research points to new model for determining surgical candidates among patients with epilepsy

American Academy of Neurology singled out work as top research project 

Dr. Zeke Gleichgerrcht and Dr. Logan Schneider
Dr. Ezequiel Gleichgerrcht, left, shakes hands with Stanford University sleep neurologist Dr. Logan Schneider, a member of the American Academy of Neurology awards committee. Photo provided 
Leslie Cantu | cantul@musc.edu | Aug. 7, 2018

More than 3 million people in the U.S. suffer from epilepsy, and only about two-thirds can control their seizures through medication. 

For some with uncontrolled epilepsy, surgery is the answer. About two-thirds of patients who undergo surgery will find relief from seizures, said MUSC neurology co-chief resident Ezequiel “Zeke” Gleichgerrcht, M.D. But that means surgery fails for one-third of patients who go under the knife. And right now, doctors can’t reliably say beforehand which group a patient will fall into. 

Gleichgerrcht is hoping to change that by coming up with a better prediction model not only to indicate which patients would benefit from surgery, but also which newly diagnosed patients would benefit from medication.

It can take up to 20 years before a patient with epilepsy is referred for surgery, as patients try one drug after another. Yet the risk of severe side effects from epilepsy, including death, is greater with the accumulation of years than the risk from surgery, Gleichgerrcht said. 

His initial research project, under the guidance of mentor Leo Bonilha, M.D., Ph.D., won him the American Academy of Neurology Alliance Founders Award this year as the top investigation by a neurology resident.

Dr. Leonardo Bonilha 
Dr. Leonardo Bonilha 

That’s because he was able to predict, with up to 90 percent accuracy, which patients would be seizure-free, with seizure-free defined as free of disabling seizures. 

To do this, Gleichgerrcht returned to an old idea that has re-emerged in the last decade with the advances in neuroimaging and computational models — that epilepsy is a disease of networks. 

Epilepsy is labeled according to a number of factors, including the part of the brain involved. For example, temporal lobe epilepsy, the most common and also best understood form of epilepsy, involves seizures that originate in the temporal lobe. 

Yet research has shown that other parts of the brain, beyond the area where the seizures originate, are permanently changed as well in patients with seizures.

“How much of that is consequence and how much of that is cause is a very hard question. It’s probably a vicious cycle,” Gleichgerrcht said. 

But these changes in other parts of the brain seem to be the reason some people continue to have seizures, even after surgery that removed what was thought to be the problematic area. 

“You’re expecting this patient to have only changes in the temporal lobe, because it’s temporal lobe epilepsy; then why are there changes in the frontal lobe, in the occipital lobe, in the parietal lobe? It’s because epilepsy is a disease of networks. More and more research is showing that these changes may be the ones that propagate seizures, beyond the focus,” he said. 

Yet the changes themselves, as well as the pattern of changes that becomes significant for maintaining seizures, aren’t discernible to the human eye. 

That’s where some math and some deep learning by machines come in. 

Gleichgerrcht went back and looked at pre-surgical diffusion MRIs for 50 patients with temporal lobe epilepsy from a five-year period. None of the patients had an obvious cause of their epilepsy, such as a tumor or a vascular malformation. 

Diffusion MRIs trace the path of water molecules along axons in the brain. Continuing the line of research pioneered by Bonilha, his research mentor, Gleichgerrcht applied graph theory, which is a way of describing complex networks, to use the information from the diffusion MRIs and create a topographical map of each brain, dividing the brain into 384 nodes of interest. 

For each patient, the “betweenness centrality” of every node in the network was then calculated. Betweenness centrality is a measure of influence in a network. It doesn’t simply depend on the number of connections each node has but how crucial they are for the network. Using a familiar network to explain the concept, “Joe” might have 500 friends on Facebook, while his college buddy “Bob” has only 100. But if Bob, the sole link between two groups of people – his college friends and his work colleagues – were to be removed from the network, there would be no way for the two groups to connect. Bob has more betweenness centrality than Joe. 

visualization of betweenness centrality

A representation of the concept of betweenness centrality. Node D has the highest degree (number of links to other nodes), but if it were to be removed from the network, information could still flow with relative efficiency. Node H has only three connections, but acts as a hub between nodes I and J and the remainder of the network. Thus, node H is the one with the highest betweenness centrality, acting as a true influencer in the network. Image provided

In looking at the betweenness centrality of nodes in the brain, Gleichgerrcht chose the top 10 influencers that seemed to be most associated with patients either becoming seizure-free or non-seizure-free. The patients who were seizure-free showed different influencers than the patients who still had seizures. The research also showed the influencers weren’t in the medial temporal area. 

This finding has opened a new line of research, Gleichgerrcht said. An important next step would be to do a multi-center study to verify the findings. What could make this so significant is that it doesn’t require new equipment or a new surgical technique. Any medical center that performs epilepsy surgery has diffusion MRI capability, he said, and patients already get this test as part of their workups. His dream would be for a prediction score to be generated automatically with the MRI report, giving the team of neurologists and radiologists another point of consideration as they discuss each patient’s suitability for surgery. 

Gleichgerrcht, who will be the epilepsy fellow in the Department of Neurology next year, is continuing the research by applying deep learning, a type of machine learning, to look for complex patterns. Although his research project focused on the top 10 influencers, with deep learning the machine can take data from all nodes and detect the importance and influence of each. He’d also like to add in different types of data. Could surgical predictions be enhanced even more if information about the structural connectome and functional connectome – the parts of the brain that are physically, structurally connected and the parts that are functionally connected – were added? What if the results of a PET scan were also added? 

“The data are there, it’s just a matter of constructing a model that is able to process that data that the human eye can’t combine,” he said.

He’d also like to look at less understood epilepsies and find a database of appropriate samples from which to generate a model of which patients will likely be able to control seizures through medication. 

 

New laser technology helps patient live her life. (MUSC Catalyst News, July 20, 2015) 

 

Suggest a story to MUSC Catalyst News



 

 

View MUSC's Facebook page Follow MUSC on Twitter View the MUSC Health Youtube channel Read the MUSC Health blog circle arrow MUSC_TAG_SOLID_1C