Description
Approximately 60 million people worldwide suffer from epileptic seizures. A key challenge in machine learning approaches for epilepsy research is the lack of a data resource of analysis-ready (no additional preprocessing is needed when using the data for developing computational methods) seizure signal datasets with associated tools for seizure data management and visualization. We introduce SeizureBank, a web-based data management and visualization system for epileptic seizures. SeizureBank comes with a built-in seizure data preparation pipeline and web-based interfaces for querying, exporting and visualizing seizure-related signal data. In this pilot study, 224 seizures from 115 patients were extracted from over one terabyte of signal data and deposited in SeizureBank. To demonstrate the value of this approach, we develop a feature-based seizure identification approach and evaluate the performance on a variety of data sources. The results can serve as a cross-dataset evaluation benchmark for future seizure identification studies.
Learning Objective: 1. A shared data resource for analysis-ready, seizure-related signal datasets.
2. A web-based data management and visualization system for epileptic seizures.
3. A cross-dataset evaluation benchmark for further seizure identification studies.
Authors:
Xiaojin Li (Presenter)
University of Texas Health Science Center at Houston
Yan Huang, University of Kentucky
Shiqiang Tao, University of Texas Health Science Center at Houston
Licong Cui, University of Texas Health Science Center at Houston
Samden Lhatoo, University of Texas Health Science Center at Houston
Guo-Qiang Zhang, University of Texas Health Science Center at Houston
Presentation Materials: