Interesting Recommendations Based on Hierarchical Visualizations of Medical Data

Abstract

Due to the dramatic growth in Electronic Health Records (EHR), many opportunities are arising for discovering new medical knowledge. Those kinds of knowledge are useful for related stakeholders such as hospital planners, data analysts, doctors, insurance companies for better patients’ management. However, many challenges need to be addressed while dealing with medical domain such as (1) how to measure the interestingness of information (2) how to visualize such interestingness (3) how to handle high dimensionality of medical data. To address these challenges, we present MedVIS, an interactive tool to visually explore the data space and finds interesting visualizations compared with another dataset. MedVIS structures visualizations into a hierarchical coherent tree to reveal interestingness of dimensions and other measures. We introduce a novel analysis work flow, and discuss various optimization mechanisms to effectively and efficiently explore the data space. Additionally, we discuss various approaches to mitigate the problem of high-dimensional medical data analysis and its visual exploration. In our experiments, we apply MedVIS to a real-world dataset and show promising visualization outcomes in terms of effectiveness and efficiency.