Interactive Visualization of Model Results and their Comparison - regarding infection control in hospitals
During a hospital visit each year about 400,000 to 600,000 patients acquire an infection in connection with a medical measure that has taken place. It is estimated that between 10,000 and 15,000 (about 2.5%) patients die each year in Germany as a result of these infections.   Furthermore the number of infections has increased drastically since 2020, because of SARS-CoV-2. Modern Machine Learning and Deep Learning in particular have achieved groundbreaking results in different fields  . For this work a Deep Learning model has been provided by our research partner NEC Laboratories Europe GmbH. The model is able to predict the infection risk for a given patient. However it is not clear how to present the complex data to the user in an understandable manner. This is the aim of this work. To answer the research question the provided data was carefully analyzed. The model results were compared with the provided input data to show conspicuous features. Furthermore Explainable AI techniques were researched in order to add interpretability to the predictions made by the model. An interactive dashboard was constructed using web technologies. The dashboard shows important features such as the time procedure for the infections for different parts of the hospital. The layout was designed using feedback from Prof. Dr. Tatiana von Landesberger who is the head of the Chair for Visualization and Visual Analytics at the University of Cologne and Tom Baumgartl a Student Assistant at Technical University of Darmstadt and University of Cologne. Both have already worked on previous work in this field. Regarding the goal to present complex data in an understandable manner, the initial reactions from those involved were consistently positive. However detailed user feedback will only be obtained shortly after submission, so no definitive answer to the success of this work can be given yet.
Darmstadt, TU, Master Thesis, 2021