In Denmark, approximately 1500 new cases of head and neck cancers are diagnosed every year. Radiation therapy is the most common treatment for this type of cancer, and shall not be taken lightly. The side effects of such treatment could decrease the life quality of a patient.
Currently, physicians are manually delineating the tumor tissue location on images from computed tomography (CT) scanning. Besides being time inefficient this method also leads to intra- and inter-observer variation.
Having this as the motivation, the focus of this project is to develop a framework that segment tumor tissues based on head and neck cancer CT scanning images. This framework uses classification-based segmentation with texture information as input. The developed framework have first been cross-validated, and from this a trained-classifier have been created. Finally, the trained-classifier have been tested on new images from two patients. The results from the cross-validation showed a sensitivity of 66.58%, specificity 98.85%, and accuracy of 95.94%. The results obtained from the test using the trained-classifier, is as follows: sensitivity 22.8%, specificity 90%, and accuracy of 64%.