3D lymphoma detection in PET-CT images with supervoxel and CRFs
Résumé
In this paper we present a lymphoma detection method on image PET-CT by combining supervoxel and conditional random fields(CRFs). Positron-emission tomography(PET) is often used to analysis diseases like cancer. And it is usually combined with computed tomography scan (CT), which provides accurate anatomical location of lesions. Most lymphoma detection in PET are based on machine learning technique which requires a large learning database. However, it is difficult to acquire such a large standard database in medical field. In our previous work, a new approach which combines an anatomical atlas obtained in CT with CRFs (Conditional Random Fields) in PET is proposed and is proved to have good results, however it is very time consuming due to the fully connection of each voxel in 3D. To cope with this problem, we proposed a method that combines supervoxel and CRFs to accelerate the progress. Our method consists of 3 steps. First, we apply the supervoxel on the PET image to group the voxels into supervoxels. Then, an anatomic atlas is applied on CT to remove the organs having hyper-fixation in PET. Finally, CRFs will detect lymphoma regions in PET. The obtained results show good performance in terms of speed and lymphoma detection.