A new similarity measure for deformable image registration based on intensity matching

Deformable image registration plays an important role in medical image analysis. Multi-modal image registration remains a challenging research topic due to the complexity of modeling the relationship between two images. Mutual information (MI) is widely used in the field of multi-modal image registration, however, it suffers from problems such as interpolation artifacts and/or statistical insufficiency. The problem is worsened when bias field and noise are present. There have been attempts to map images to a common modality before image registration process, but the error introduced by the mapping may be detrimental to the registration. In this paper, instead of explicitly mapping the images to a common modality, we introduce a new similarity measure based on intensity matching information, which can be learnt from the existing registered training pairs or images pairs registered by performing MI based registration. Experiments on simulated brain MRI and real myocardial perfusion MR image sequences indicate that our proposed similarity measure outperforms the conventional MI and Kroon and Slump's method [1].