Automatic mesh morphing of whole bone finite element models to clinical CT data

This project will be part of an EPSRC funded PhD network (three projects in total) where the candidates will form, together with the supervisors and co-supervisors, a research group focused on the development and improvement of current elastic registration methods in musculoskeletal applications. The network will be part of the INSIGNEO institute for in silico medicine (http://insigneo.org/).
Insigneo controls one of the most refined patient-specific modelling protocols in the world to predict the strength of patient’s bones simply from appropriately calibrated clinical computed tomography (CT) images, of the patient’s skeleton. This fully validated protocol is used by our researchers to investigate osteoporosis, bone metastases, back pain, osteogenesis imperfecta, identification of paediatric abuse fractures, juvenile idiopathic arthritis, knee replacement, etc.

Unfortunately, the protocol requires approximately one day’s work by a skilled operator to transform the patient’s images into a finite element model capable to accurately predicting bone strength. Most of the operator’s time is spent segmenting the bone geometry from CT images, and then decomposing this geometry into a well-conditioned finite element mesh. In the course of the project VPHOP, which we coordinated, our partner Ansys developed a mesh-morphing algorithm that could “fit” a generic finite element mesh to the bone geometry of each patient. This provided the necessary automation, and added a convectional partitioning of the anatomical space that could be used for multiscale modelling. However, the Ansys technology still requires the manual segmentation of the images; this accounts for about half of the operator’s time. Furthermore, the resulting mesh is not optimal: in a recent study we found that by using this mesh morphing as opposed to the manual protocol the ability of our patient-specific models to classify fractured patients accurately decreased by nearly 10%.

The Sheffield Image Registration Toolkit (ShIRT v2) computes the registration transformation in parametric space. This means that we can a) directly register finite element meshes onto 3D images, and b) add to the regularisation term mesh conditioning metrics, so that the registration preserves the optimal mesh conditioning of the template mesh. This method has already been demonstrated in a similar problem in cardiac modelling, but has not as yet been used for bone modelling. As a further advancement, we will explore how the regularisation term can be used to improve the handling of the tissue property heterogeneity over the integration domain.