A multimodal approach towards epileptic seizure detection and prediction

点击翻译按钮获取中文摘要

Epilepsy is one of the most common neurological diseases with far-reaching consequences for patients. Patients describe the uncertainty about when an epileptic seizure will occur as the greatest burden in their everyday lives. The disease itself results in recurrent epileptic seizures, which originate from excessive electrical discharges from groups of neurons and have been shown to alter the activity of the autonomic nervous system (ANS). Multimodal wristbands offer a non-invasive and easy-to-use possibility to continuously monitor ANS functioning. However, biomarkers and seizure prediction algorithm still lack specificity due to the high inter- and intra-individual variability present in ANS activity. The overall goal of my project is to identify biomarkers for seizure susceptibility and to develop a seizure prediction algorithm. I hypothesize that seizure-induced alterations of central activity, which show themselves within and across subsystems of the ANS, follow a multimodal pattern. Related to hyper-excitability of the brain in pre-ictal phases, which potentially alters the central drive of the subsystems of the ANS, such as cardiac, respiratory, and electrodermal system, I expect multimodal markers that express the amount of information shared between subsystems to increase.To analyze multimodal ANS changes, I will analyze recordings of continuous heart rate, respiratory rate, electrodermal activity and peripheral body temperature measured at the wrist while seizures were recorded and classified based on continuous video-EEG recording. To test theory-inspired biomarkers I will characterize the unimodal signals information content by use of entropy measures within subsystems and the information exchange by use of mutual information measures. To develop a seizure prediction algorithm that can identify high risk for seizure in ANS data. By use of uni- and multimodal biomarkers that indicate seizure susceptibility will be used to develop classifiers and neural networks that allow for seizure prediction on an individual and a group level first for GTCS only and then extend it to all seizure types.The project is highly feasible due an existing working relationship and availability of data. The host institute has a unique database of approximately 300 ANS recordings of children with diagnosed epilepsy. The expected outcome are biomarkers that are relevant across patients and allow for a reliable individualized seizure prediction in most patients. The results are expected to help improving the quality of life in PWE, their relatives and caregivers, and possibly improve controllability of seizures and reduce health care costs.