The thesis deals with the development of a visualization tool for physicians to display data sets of stroke patients in 3D. The aim is to enable an improved visualization of medical research results.
First, the medical phenomenon of stroke is explained. Then a deep learning system is presented that uses neural networks to predict stroke lesions. The underlying data of this system – CT scans, medical ground truth annotations and the predictions of a convolutional neural network (CNN) – form the basis of the work.
This is followed by a presentation of various methods of 3D visualization and the visualization of uncertainties in medical data sets. Based on this, user-related prerequisites and functional requirements for the application are formulated. Before implementation, suitable software solutions for developing the application are analyzed.
The implementation is carried out in two phases. Particular attention is paid to the representation of uncertainties, whereby various techniques for visualizing these uncertainties are described. Finally, the developed application is evaluated, its limitations are pointed out and possible further developments are discussed.