Groupwise image registration of multimodal head-and-neck images
Fusion of multimodal medical images using deformable registration is of high interest for head-and-neck tumour treatment planning. In this context, more than two images often have to be aligned for a given patient. The conventional, pairwise way to register multiple images is to select one of them as fixed reference and independently align each remaining image with it. An alternative method would be to simultaneously register the images using a groupwise registration scheme, thus eliminating the need to select a reference image and avoiding any bias due to this arbitrary choice. In this study, we propose a novel groupwise image registration technique, combining a principal component analysis (PCA) based similarity metric and modality independent neighbourhood descriptors (MIND). Results on 16 patients show that the images are slightly better aligned when using the proposed registration method than when using pairwise registration.