Epilepsy Bioinformatics Study for Antiepileptogenic Therapy (EpiBioS4R)

The Epilepsy Bioinformatics Study for Antiepileptogenic Therapy (EpiBioS4R) Scientific Premise is: epileptogenesis after traumatic brain injury (TBI) can be prevented with specific treatments; the identification of relevant biomarkers and performance of rigorous preclinical trials will permit the future design and performance of economically feasible full-scale clinical trials of antiepileptogenic therapies.

The ICON Lab is specifically contributing to the EpiBioS4Rx Consortium by identifying multimodal biomarkers of epileptogenesis. We apply data science, mathematics, and machine learning to glean insight into what factors may induce the development of seizures after brain injury so we can ultimately predict which patients will develop epilepsy. Our work centers around biomarker identification in scalp and depth electrophysiology and multimodal imaging data.

Automating Lesion Segmentation 

TBI patients often have significant brain deformations due to hemorrhagic and contusional injuries. These deformations impact downstream imaging analysis by disrupting visual landmarks and voxel properties used to determine tissue class and perform registration. Therefore, a clear map of affected brain tissue is essential for accurate analysis. However, there is no toolbox for lesion segmentation of TBI patients, so we are performing manual segmentation for hundreds of patients to establish ground truths to train machine learning classifers to perform automated lesion segmentation. Individual lesion masks can be stacked to evaluate the degree of overlap across the EpiBioS4Rx population.

Exploring How Lesion Characteristics Influence Seizure Development 

Once lesion masks are generated, we can use them to evaluate how specific characteristics of the injury (e.g. location and volume) relate to developing post-traumatic epilepsy.

Improving High Frequency Oscillation Detection 

High frequency oscillations are one potentially pathogenic pattern found in electroencephalography that may distinguish TBI patients who develop subsequent epilepsy. Although insightful, HFOs are difficult to detect because they are brief and transient. Manual detection is tedious and time-intensive, and existing detection algorithms yield high false positive rates. Therefore, we are working towards developing a machine-learning based classifier that can improve the sensitivity of HFO detection.

Once HFOs are identified, we are using them to build predictive models of HFO progression in rodents after induced injury.

This work is supported by the National Institute of Neurological Disorders and Stroke (NINDS) at the National Institutes of Health (NIH), award numbers U54NS100064  and R01NS111744.

For more information, please visit EpiBios4Rx site.