Speaker
Description
Compared to the core MIMIC databases, the MIMIC waveform databases are underutilized, even though they contain valuable, openly available intensive care unit bedside monitoring data. This disparity may be driven by the Waveform Database (wfdb) format, which may be unfamiliar to some researchers. Typically, preprocessing steps are applied to waveform datasets before they can be used for data analysis or machine learning. We propose WavePrep, an open-source tool that takes a MIMIC waveform database, selects user-specified records and time spans, applies configurable preprocessing steps to them, and produces a CSV dataset ready for further research. The tool lowers the entry barrier to physiological signal research by abstracting from preprocessing logic and parallelization, allowing the user to specify channel-specific preprocessing steps in a user-friendly JSON format.