[home] [Personal Program] [Help]

---

Session: Image Analysis - Cancer
Session starts: Friday 25 January, 13:00
Presentation starts: 13:45
Room: Lecture room 559

---

Valerio Fortunati (Biomedical Imaging Group of Rotterdam, Department of Medical Informatics and Radiology, ErasmusMC, Rotterdam, The Netherlands)
Rene Verhaart (Hyperthermia Unit, Department of Radiation Oncology, Daniel den Hoed Cancer Center, Erasmus MC, Rotterdam, The Netherlands)
Wiro Niessen ()
Jifke Veenland ()
Maarten Paulides ()
Gerard van Rhoon ()
Theo van Walsum ()

Abstract:
Introduction Optimization of hyperthermia treatment of head and neck tumors requires accurate treatment planning. Hyperthermia treatment planning (HTP) is based on tissue segmentation for 3D patient model generation. This process is currently performed manually, which is a tedious and time consuming procedure limiting the clinical applicability of this treatment. Methods We developed and evaluated a fully automatic multi atlas-based segmentation algorithm for the temperature sensitive (critical) tissues of head and neck CT images. These are: Cerebrum, Cerebellum, Brain-Stem, Myleum (Spinal-Cord) [1] and Eyeballs [2]. To overcome the large anatomical variability, multi atlas registration and intensity-based classification were combined. Each tissue is segmented independently using binary graph cut method [3]. A cost function composed of 1) an intensity energy term, 2) a spatial prior energy term based on the atlas registration and 3) a regularization term is globally minimized using graph cut. Afterwards the results from each tissue are combined. Experiments and results Data was collected from patients scheduled for combined radiotherapy and hyperthermia (HT) treatment. The data used in the evaluation study comprised 18 axial CT images scanned with the same protocol. These images were manually delineated from a medical radiation technologist student which was trained by radiation oncologist. Using this dataset the method was tuned and evaluated in a nested leave-one out experiment. The contribution of adding intensity based classification and regularization to atlas registration is evaluated comparing the graph-cut method (GC) with majority voting combination (MV) [4] of the multi atlas registration results. Both GC and MV results are evaluated with respect to manual delineations used as ground-truth. For the evaluation Dice similarity coefficient (DSC) and mean surface distances (MSD) were calculated. Combining intensity classification, regularization and multi atlas registration, the accuracy of the segmentation of Cerebrum, Cerebellum and Eyeball is significantly improved (tested using two sided, paired, Mann-Whitney test, p>0.05) with respect to MV. Overall a high correspondence was found for these tissues with a DSC median value higher than 0.87 and MSD lower than the voxel resolution in-plane (1mm). The method does not give a significant improving for Myleum and Brain-Stem segmentation accuracy. While the Myleum accuracy is comparable with the other tissues results in DSC and MSD, the Brain-Stem segmentation is less accurate. Both tissues are hard to be discerned from their surrounding relying on intensity information and this may be the cause of failure in accuracy improvement. Conclusion We are planning to investigate the inter-observer variability in method accuracy using different atlases for segmentation. Further the influences of the variation of accuracy on the SAR and temperature distribution prediction (and so on HTP) will be analyzed. The proposed method has potential to improve the efficiency of tissues delineation for HTP. REFERENCES [1] P. Sminia, J. van der Zees, J. Wondergem, J. Haveman, “Effect of hyperthermia on the central nervous system” Int. J. Hyperthermia, 10(1), 1-30. (1994). [2] J.A. Elder, “Ocular Effects of Radiofrequency Energy”, Bioelectromagnetics Supplement, 6, 148-161, (2003). [3] F. van der Lijn, T. den Heijer, M.M.B. Breteler, W.J. Niessen, “Hippocampus segmentation in MR images using atlas registration, voxel classification, and graph cuts” Neuroimage, 43, 708-720, (2008). [4] S. L. Hartmann, M. H. Parks, P. R. Martin, and B. M. Dawant, “Automatic segmentation of internal structures of the head in MR images using a combination of similarity and free-form transformations: Part II, validation on severely atrophied brains IEEE transactions on medical imaging 18, 917 (1999).