Ontology-Based Medical Information Extraction through Machine Learning for Disease History Analysis

Visual Analytics has developed many new and exciting ways to transform and visualize data to solve complex analytical tasks. Therefore, Visual Analytics is applied in various domains. Each application requires different data transformation models and visualization to fit users' demands. A significant application area with complex and unstructured data is healthcare. In this context, the course of patients' disease plays an increasing role since developed countries already have sufficient data on patients' diseases, medications, and therapies. However, a medical doctor has a concise time replicating the entire disease course. The decision-making process relies on verbal communication combined with a short view of the primarily digital patient file, which does not allow for investigation of the entire patient's history of disease and making sufficient decisions on the follow-up therapies. Visual Analytics that combines machine learning methods and interactive visualization could provide a single visual history of disease along with medication and other health metrics, e.g., blood values, weight, and increasing or decreasing disease count temporally and semantically. Data using machine learning data analytic approaches, we will employ clinical data from patients' courses of diseases to support medical doctors in accurately identifying the disease and the best way of treatment. The visual support illustrates the history of therapy and diseases for quick and reliable identification of possible diseases and treatments. Electronic health records (EHRs) from a collaboration partner in Germany are the baseline of the learning models. The related data are commonly unstructured text data and differ in wording and content. By applying the learning methods that retrieve named entities, such as disease names, medication, and treatments, in combination with socio-demographic data and side effects of drugs, medical doctors will be able to identify the disease more reliably quickly. This thesis introduces a visual analytics system that transforms unstructured text data by taking advantage of the most common EHR categorizations to support medical doctors in identifying and treating patients in the best possible way.