574: Enhancing the Interoperability and Ai Readiness of Neurocritical Care Data

Abstract

INTRODUCTION:Artificial Intelligence (AI) technologies have the potential to greatly enhance our use of neurocritical care (NCC) data to provide precision management of brain injury patients. Work in this area aligns with the NIH Big Data to Knowledge initiative. However, barriers exist that prevent widespread use of AI with NCC data, including lack of standardization for data labels, monitoring procedures, and device connectivity as well as the need for high-resolution data. With funding from the US Department of Defense, we addressed these barriers to enhance the use of AI for NCC data. We report here our further progress in developing a standardized data infrastructure and specifically our work in making the data AI-ready. METHODS:We focused on four areas: data integration, data harmonization/annotation, standard archive formats, and a curation/analytics dashboard. We developed cloud connectivity to upload data from commercial device integrators, electronic health records, and image and genomic repositories. We developed an ontology that provides a knowledge structure for NCC data that was used for harmonization. From annotations collected from the TRACK-TBI data set, we created a standardized annotation format. We converted all data into a standardized format (HDF5) and created an application programming interface (API) to use this data. A tool was developed to create, sequence, and share interoperable modules through a user-friendly graphical user interface. A web-based dashboard was built to control all functions and allow for visualization of integrated data such as time-synchronized physiology and medications. RESULTS:The features and use of these tools have been validated by their adoption in clinical trials (BOOST-3/ ELECTRO-BOOST, TRACK-TBI, and PRECICECAP). The API and modular application concepts were successfully tested in an AI-based seizure detection study. CONCLUSIONS:Our dashboard and data infrastructure have been successfully used in several clinical trials and preliminary AI and machine learning explorations. Our work to date has focused on post-hoc data processing and analytics. This work has led to foundational technological advancements that will lead to real-time continuous precision management of brain-injured patients.