Evaluation of inter-individual pelvic CT-scans registration.
Résumé
Image registration appears to be an essential tool in a wide range of biomedical applications. We are interested in performing voxel-wise population analysis aimed at explaining sideways effects of dose irradiation across a population treated for prostate cancer. To this end, a perfect registration of pelvic computed tomography (CT) scans is required allowing a reliable mapping of dose distributions in a single coordinate system. In this paper, we compared the performances of four different inter-individual intensity-based registration strategies: two affine (block matching and mask-restricted block matching) and two non-rigid (diffeomorphic demons and free-form deformation). Pelvic CT-scans were registered towards a single template in a leave-one-out cross validation scheme. The ability of the methods to align inter-individual images was assessed in terms of overlap measures between the manually segmented organs after the registration. Results show that the free-form deformation registration strategy outperformed the other methods and provided more accurate registrations. We showed that precision can be improved by restricting the algorithm calculations to the main regions of interest.