Segmentierung von 3D-MRI-Volumendatensätzen

Purpose This work addresses design and implementation of an application for segmenting medical volume data. The software to be implemented (MIP-Medical Imaging Platform) is designed as a platform for developing and testing general medical image processing applications. To segment and analyze time series of three-dimensional nuclear magnetic resonance images of the human heart, with regard to various cardiological characteristics, is considered as a typical use case. Based on the observations gained from examining this specific application, primary product requirements are determined. Methods and Tools Planning and design of the application are carried out according to common principles of Software Engineering. After specifying the product requirements, an object-oriented analysis model is developed. This model is refined to form the system design which is finally implemented in C++. Several software libraries are employed: The toolkit ITK (Insight Segmentation & Registration Toolkit) provides a wide range of image processing functionality. Visualization is addressed using VTK (Visualization Toolkit). The framework wxWindows is used to implement the graphical user interface. Results In order to ensure easy extensibility for new areas of application, various approaches of how to efficiently organize and manage image data are examined. Since a multitude of image types is commonly used in medical image processing, these alternatives are particularly tested for the type flexibility they provide. The approach which emerges from this process is reflected in the whole application, which can be controlled entirely by means of type-independent interfaces. Not only does this apply to the management of arbitrary image types, but also to the handling of different image processing procedures. The use of object-oriented design patterns helps to realize this approach in a convenient way. Image processing usually comprises consecutive filtering of image data. For this purpose, filter chains are established, where the output of one filter serves as the input of its successor. While ITK already basically provides this functionality, it is extended in order to enable consistent processing of series of images. Filters are encapsulated in classes and can be parameterized via a common, type-independent interface. Two segmentation methods-Region Growing Connected Threshold and Geodesic Active Contours-are implemented and applied to authentic medical data. The graphical user interface is designed to satisfy the developer's rather than the physician's needs and thus provides full control over image processing activities. Dialogs for configuring filters and controlling the data flow are generated automatically, as all information required for this can be retrieved by querying consistent interfaces of the respective class hierarchy. Conclusion The designed software is intended to serve as a foundation for developing and examining a wide range of applications in medical image processing. This purpose is promoted by consistent data flow handling and the provided image type flexibility. Commonly used features like visualization and file handling, as well as the graphical user interface, have been implemented with regard to possible future extensions