Hypoxia is a marker of poor prognosis in malignant tumors independent from the selected therapeutic method and the therapy should be intensified in such tumors. Hypoxia imaging with positron emission tomography (PET) is limited by low contrast to noise ratios with every available tracer. In radiation oncology appropriate delineation is required to allow therapy and intensification. While manual segmentation results are highly dependent from experience and observers condition (high inter- and intra observer variability), threshold- and gradient-based algorithms for automatic segmentation frequently fail in low contrast data sets. Likewise, calibration of these algorithms using phantoms is not useful. Complex computational models such as swarm intelligence-based algorithms are promising tools for optimized segmentation results and allow observer independent interpretation of multimodal and multidimensional imaging data.