Medical Image Registration

One of the major issues in medical imaging is the analysis and processing of multiple images in a common frame of reference. Placement in a common frame of reference, known as registration or spatial mapping or spatial normalization, or sometimes image fusion, enables the spatial superposition of anatomical or functional structures present in each of the images. The aim of this article is to provide the reader with an overview of the various aspects of medical image mapping.

Registration is an essential step in many medical image processing chains, and its applications are numerous. Two application contexts can be distinguished: intra-individual registration, which involves images from the same subject, and inter-individual registration, which matches images from different subjects or from a subject and a reference model, commonly called an atlas. Intra-individual registration generally requires the determination of a simple geometric transformation, characterized by a small number of parameters, such as a rigid or affine transformation. It is notably used to merge images from different modalities, to detect and monitor the evolution of lesions or tumors, and to measure geometric modifications of certain anatomical structures (atrophy, hypertrophy, displacement, monitoring and characterization of cardiac and respiratory movements, etc.). Inter-individual registration often requires the estimation of transformations described as "non-rigid" or "deformable", with a much greater number of degrees of freedom. Its main applications are in the construction and use of probabilistic atlases, for example for automatic segmentation by transport of an a priori model, or for the statistical analysis of anatomical or functional variations, normal or pathological, within a group of individuals. Registration can also be used to correct geometric distortions in images.

The nature of the images to be matched can be very diverse, and greatly conditions the choice of the matching method to be used. It is possible to match images from the same modality (monomodal registration) or from different modalities (multimodal registration). The nature of the modalities considered (MRI, nuclear medicine, ultrasound, X-ray...