Next: Atlases and Deformable Models Up: Title and abstract Previous: Title and abstract

Introduction

In surgery planning it is often desirable to match, or register different modality datasets to make the best use of image information. In image guided surgery the aim is to register the preoperative images to the patient in the operating room. For both these purposes the conventional method is to assume that a rigid transformation is applicable. Since this does not take into account any tissue deformation these procedures are limited to rigid regions of the body, such as the skull base. We propose a physical model for non-rigid registration based on a grid of spring-like elements. Different tissues are labelled according to their physical properties, which translate to spring constants in the model. For initial model we label tissues as either rigid, deformable or fluid. This model can then be deformed using either landmark data (points) or higher dimensional data from intraoperative images. Our model requires segmentation of different tissue types. The visible human dataset (VHD) could provide a useful atlas for this process. The physical properties of tissues together with boundary conditions are labelled in the VHD. The problem then becomes one of finding a match between the VHD and the individual. Several algorithms have been proposed for this purpose 1,2. We intend to use a multi-scale approach based on locally rigid regions 2.

The VHD could also be used to simulate the process of intraoperative imaging. For example a slice through the dataset could be blurred and have noise and speckle added to simulate an ultrasound scan. We could then use this synthetic scan to compare algorithms for matching ultrasound to preoperative scans. Since we have the high resolution data available we can accurately examine the effect that noise and resolution have on these algorithms. Simulation of deformation is another possible application. If we apply our physical model, or perhaps a more complex one, to the high resolution data, a physically plausible deformation of tissue could be created. Warped MRI and CT can then be produced, either synthetically or by resampling the already registered datasets. This information might then be used to examine how algorithms, such as our simple physical model, perform in registering images and predicting global tissue deformation.

Next: Atlases and Deformable Models Up: Title and abstract Previous: Title and abstract