Brief description

In modern oncology, non-invasive tomographic imaging technologies have a growing importance. The Biological and Molecular Imaging group of OncoRay is working on development and validation of clinical imaging techniques that visualize and enable quantification of biological processes of tumours in vivo. An issue for computer scientists working in this field is the development of accurate segmentation algorithms for target volume delineation and volume analysis. In this project an automatic segmentation algorithm based on ant colony optimization for low contrast positron emission tomography (PET) is developed. Especially when analysing low contrast PET data sets the delineation of a target volume is not trivial (see figure 1). After first promising results of segmentation of phantom data sets (see figure 2) the approach is being further developed to be applied on patient data sets. Current investigation is about

• What is the minimum contrast keeping the algorithm successful in segmenting data sets?
• What is the minimum size of a target object which can be recognized by the approach?
• How can the approach be applied to delineate target objects with structures as expected in a biological environment?
• How is performance enhancement achievable by code optimization and algorithm parallelization?
• How can variability, reproducibility and repeatability of the algorithm be improved to make clinical interpretation possible?
• What image data from other imaging modalities (computed tomography, magnetic resonance) must be included in the segmentation process to improve delineation quality?