Motion uncertainty visualization based on Gaussian process regression

In the context of image-guided soft-tissue intervention and surgery navigation, soft tissue motion estimation for a deformed/registered pre-operative/pre-interventional medical image may suffer from multiple sources of uncertainty. Visualizing the uncertainty in the deformed image is of great interest because it conveys the uncertainty in the registered pre-interventional/pre-operative navigation information on deformable soft-tissue organs governing the decisions of the surgeon or intervention specialist. We address the problem of propagating the uncertainty from a Gaussian process (GP)-based motion estimation of an input image. To visualize the deformed image itself, we show its maximum a posteriori (MAP) estimate. To convey the overall uncertainty in the MAP deformed image estimate, we apply a spatially-varying blur (according to the uncertain motion estimate) to an image containing a grid and combine it with the MAP deformed image itself. We show 2D cross sections of the final image that convey both the deformed image and its uncertainty. We further extend our approach to local investigations of points or regions of interest (e.g., a tumor). To visualize where a point/region of interest (POI/ROI) is after deformation, we estimate marginal (i.e., per voxel) label (e.g., for being a tumor) probabilities (MLPs) based on an input probabilistic segmentation of the ROI in the non-deformed pre-interventional/pre-operative image and the input GP posterior distribution over deformation fields and show 2D cross sections of visually mapped MLPs or MLP isocontours.