A graduate student in Semel Institute for Neuroscience and Human Behavior, and two of his undergraduate trainees, were selected by the Council on Undergraduate Research to present their original research on Capitol Hill in Washington, D.C., on April 29.
Wesley Kerr, whose graduate studies will be completed in the UCLA Department of Biomathematics, and his trainees, Akash Patel and Sarah Barritt, will present their poster entitled "Computer-Aided Diagnosis of Epilepsy Using Clinical Information" outlining advances in the use of computational machine learning to help clinicians detect and diagnose epilepsy correctly.
"A large fraction of patients with seizure disorder are misdiagnosed and treated inappropriately," Cohen said. "As such treatments carry their own risks and negative side effects, their research could positively affect the lives of millions of Americans."
The goal of Kerr's team was to estimate the probability of epilepsy versus non-epileptic seizures based on the historical factors reported by the patient to their neurologist.
Distinguishing between epileptic and non-epileptic seizures is a challenge, Kerr said. On average, the time from the first seizure to the diagnosis of non-epileptic seizures is seven years. In the meantime, a majority of those patients are misdiagnosed with epilepsy and treated inappropriately with anti-epileptic medications.
"This can expose patients to serious, and potentially fatal, side effects," Kerr said. "One of our laboratory's goals is to create an automated system that can aid physicians in distinguishing patients with epileptic and non-epileptic seizures."
Kerr and his team accomplished this by inspecting outpatient clinical notes from patients with medication-resistant seizure disorder, who were later diagnosed as having epilepsy or non-epileptic seizure disorder, using the gold standard diagnostic assessment, 72- hour in-patient closed circuit video-electroencephalography (VEEG) monitoring.
Using a combination of the known risk factors for epilepsy and non-epileptic seizures reported in 228 clinical notes the team examined, their algorithm achieved a diagnosis accuracy of 65%. While at first glance this may appear low, it is comparable to the accuracy of neurologists prior to VEEG monitoring. In this work, they used a machine learning method known as a "decision tree."
"The structure of our decision tree also provided meaningful information about the interpretation of each risk factor in each patient. For example, the risk factors for non-epileptic seizures may not be the same for women and men," Kerr said. "This work may help diagnose, and thereby more effectively treat, patients that are in need."
In addition to serving patients with seizure disorder, the computer-aided diagnostic methods developed by Kerr and his team may be applicable to the diagnosis of other maladies in the future.
Read more here
No comments:
Post a Comment