Image guided radiotherapy is selective in the targeting of cancerous tissue. An automatic detailed delineation of the clinical target volume may decrease the treatment planning and inter-variability among clinicians. The aim of this work is to develop a vessel segmentation method to indirectly detect lymph nodes with poor visual appearance in medical images.
The method is designed with three stages. First stage uses a region growing method to extract an initial vessel tree volume. Stage two use elliptic vessel cross section features to define tube like vessel segments originally developed for image-guided peripheral bronchoscopy. Stage three connects the vessel tree volume from stage one and the vessel segments from stage two. The method is semiautomatic and only requires one manual selected seed voxel.
Fourteen bolus tracked CTA data sets were used in the test. The initial segmentation method in stage one was able to automatically select a threshold and extract an initial vessel tree without over segmentation. In addition a bone filter successfully removed bone structure. The full segmentation method was able to extend the initial vessel tree with the defined vessel segments from stage two. However, the final extracted vessel volume was prone to over segmentation, which may be caused by selected computational delimitations in the method design. Several parameters in the method enable optimisation and since the quality of the CTA data varies, the potential of the method is not fully elaborated.
In conclusion the segmentation method shows beneficial results and may potentially improve the extraction of vessel volume to guide radiotherapy.