Ontology-Based Data Preparation in Healthcare: The Case of the AMD-STITCH Project

In the context of healthcare, an AI solution is generally developed for a specific analysis task, based on a relevant dataset, with little attention to reusability and generalizability of its data preparation step. This paper focuses on a different scenario, which can be called context-oriented, where a set of clinical data sources, relevant for a specific context (e.g., a particular disease), is available and can be used for a variety of data analytics tasks, often carried out by different research groups. Therefore, the aim of this research is to present a systematic method, which exploits the Ontology-based Data Management paradigm to enhance data preparation in a context-oriented scenario. The introduced methodology has been applied to a project dealing with big data and regarding the treatment of diabetes and its complications. The peculiarity and challenge of this project lies in the fact that it deals with real world data, extracted from Electronic Medical Records within a 13 years timeframe, and thus not collected for research purposes. The paper focuses on two main steps of data preparation, namely data modeling and data cleaning, and it shows how this approach provides effective techniques for setting up a unified and shared database, to be used in the subsequent data analytics phases as an asset.