Developing tools for hyperthermia treatment planning

This project is carried out in close collaboration with the Hyperthermia Unit of the Department of Radiation Oncology (

The effectiveness of radiotherapy in the head and neck region is significantly enhanced when combined with hyperthermia. Recently, a novel hyperthermia applicator was developed for precise hyperthermia treatment of head and neck tumors. Hyperthermia treatment planning (HTP) is needed to control, monitor For treatment planning two steps are needed. First an electromagnetic (EM) model is derived from a CT-scan, this model is then used as input for the thermal model.

Until now, patient specific EM models are generated by manual outlining the different tissue types. By integrating MRI, for more-accurate tumor and soft-tissue delineation, and by introducing sophisticated registration techniques and segmentation algorithms, we can replace most of the manual operations and facilitate the fast, accurate and reproducible generation of patient specific EM models.

For the patient specific thermal model, tools will have to be developed to derive vascular tissue characteristics from MR images.

The aim of the project is to facilitate accurate and efficient model generation for developing patient-specific electromagnetic and thermal hyperthermia treatment planning for head and neck cancer patients.

The first part of the project was focused on the accurate segmentation of the thermo-sensitive organs at risk (OAR), for which the treatment temperature level has to be restricted, i.e. brain, spinal cord and eyes. Exposure of the OAR must be kept below the threshold levels to avoid toxicity of the treatment. Using hyperthermia treatment planning (HTP), the ratio between tumor dose and thermal dose deposited in the OAR is optimized. Hence, the maximum thermal dose in OARs determines directly the possibility to achieve high temperatures in the tumor. Accurate segmentation of OARs is therefore a critical issue in HTP.

An atlas based segmentation algorithm is combined with intensity models of the structures of interest. A cost function composed of an intensity energy term, a spatial prior energy term based on the atlas registration and a regularization term is globally minimized using graph cut. The method was evaluated by measuring Dice similarity coefficient, mean and Hausdorff surface distances with respect to manual delineation. Overall a high correspondence was found with Dice similarity coefficient (DSC) and mean distance values show that the overall segmentation correspondence is high for all the tissues: average DSC is higher than 0.86 for all the tissues and the average mean distance indicates accuracy within the voxel resolution.