Fuzzy tissue classification for non-linear patient-specific biomechanical models forwhole-body image registration

Abstract

Comparison of whole-body medical images acquired for a given patient at different times is important for diagnosis, treatment assessment and surgery planning. Prior to comparison, the images need to be registered (aligned) as changes in the patient’s posture and other factors associated with skeletal motion and deformations of organs/tissues lead to differences between the images. For whole-body images, such differences are large, which poses challenges for traditionally used registration methods that rely solely on image processing techniques. Therefore, in our previous studies, we successfully applied image registration using patient-specific biomechanical models in which predicting deformations of organs/tissues is treated as a non-linear problem of computational mechanics. Constructing such models tends to be time-consuming as it involves tedious image segmentation which divides images into non-overlapping constituents with different material properties. To eliminate segmentation, we propose Fuzzy C-Means (FCM) classification to assign material properties at the integration points of a finite element mesh. In this study, we present an application of the FCM tissue classification algorithm and analyse sensitivity of the accuracy of whole-body image registration using non-linear patient-specific finite models to the FCM classification parameters. We show that accurate registration (within two times of the image voxel size) can be achieved.