Multi-atlas Based Segmentation : Application to the Head and Neck Region for Radiotherapy Planning

Radiotherapy treament planning implies to delineate on the CT image of the patient the organs at risk where the dose has to be controlled. To avoid manual contouring that is tedious and prone to inter-expert variability, algorithms able to provide these delineations automatically can be helpful for the clinicians. This paper presents a multiatlas based segmentation procedure to segment the parotid glands in the context of the Head And Neck Auto-Segmentation Challenge 2010. In this procedure, the images of the training database and their manual segmentations are first resampled on the query image through the intermediate coordinate system of an average atlas. Then, a voxel-wise combination strategy based on a weighted majority vote rule is performed to estimate the segmentation of the query. For each voxel, the weight assigned to a given resampled segmentation reflects the local degree of similarity between the corresponding resampled image and the query image. We applied our method with the database provided by the workshop.