GraphXplore: Visual exploration and accessible preprocessing of medical data

Abstract

Data-driven medical research requires explainable, robust models that can handle the noisy, high-dimensional nature of electronic healthcare data while adequately communicating the results to physicians. Additionally, metadata sharing, and reproducible dataset preparation are needed to increase data quality and interoperability of privacy-sensitive patient data. In this work, we present GraphXplore, a tool for visual data exploration, automatic metadata extraction and data transformation. It enables explainable, easy-to-use exploratory data analysis paired with dataset preparation and metadata annotation accessible for physicians. The tool is implemented as a Python package and graphical user interface application tailored to the needs of medical researchers and modularly integrable into hospital data warehouse infrastructures.