Abstracts

Rationale:
Electroencephalography (EEG) is commonly used to study seizures and, in the ICU, neurological status. With advances in machine learning, computational neuroscience, and data science there is an increased interest in developing high-quality and comprehensive repositories of annotated, curated, intra- and extracranial EEG. The volume of data needed to apply computational techniques usually is large, but extracting data from proprietary systems can be prohibitively slow and cumbersome. Furthermore, the classification of patient populations requires separate clinical databases of imaging data, test results, and queries of electronic medical record systems.
Method:
In this study, we describe a new EEG extraction pipeline that aims to reduce the manual work of extracting EEG from clinical systems.  We describe the development of a series of software applications to allow the extraction of EEG data from clinical databases, bypassing clinical systems, and an automated indexing capability using clinical annotations produced by clinicians during routine clinical care.
Results:
The first step employs a de-identify tool that removes patient identifiers from the EEG data and sends the data to a HIPPA compliant REDCap database that creates a unique identifier. The second component converts electrophysiologic data from proprietary formats and extracts clinical annotations. The third program uses annotations of the CSV files and indexes the data using an Elasticsearch webserver. With this pipeline, it is possible to automate the entire process of transferring EEG in a clinical data repository to a useful and scalable de-identified research repository. This data is organized on a server that can run containerized applications for machine learning and other tasks. Furthermore, investigators can search the repository for EEGs of interest-based on the clinical annotations generated during routine clinical care.
Conclusion:
Due to the challenge of doing computational and machine learning research with clinical systems. A new automated EEG extraction and self-indexed research repository are described.  With this pipeline, it is possible to cut the time of extraction of EEG for scientific research.
Funding:
:No funding