Resumen de Advanced interaction techniques for medical models
Eva Monclus Lahoya
Advances in Medical Visualization allows the analysis of anatomical structures with the use of 3D models reconstructed from a stack of intensity-based images acquired through different techniques, being Computerized Tomographic (CT) modality one of the most common. A general medical volume graphics application usually includes an exploration task which is sometimes preceded by an analysis process where the anatomical structures of interest are first identified. The main objective of this thesis is the improvement of the user experience in the analysis and exploration of medical datasets. This improvement involves the development of efficient algorithms designed both under a user-centered perspective and taking the new computing capabilities into account in order to obtain high quality results in real-time. On the analysis stage, we have focused on the identification of the bones at joints, which is particularly challenging because the bones are very close to each other and their boundaries become ambiguous in CT images. We have concentrated our efforts on reaching maximum automation of the overall process. The proposed algorithm uses an example mesh of the same bone that has to be segmented, usually from a different person, to drive the segmentation process. The algorithm is based on an energy minimization scheme to deform the initial example mesh while following the well-defined features of the volume data to be segmented in a local and adaptive way. We also present contributions on three different aspects of the exploration task: a best-view determination system and centering in virtual reality environments, a focus-and-context technique and a point selection method. In medical practice it would often be very useful to have access to a quick pre-visualization of the involved medical dataset. We have proposed a new system which allows users to obtain a set of representative views in a short time and permits the generation of inspection paths at almost no extra cost. The technique relies on the use of a multiscale entropy measure for the generation of good viewpoints and uses a complexity-based metric, the normalized compression distance, for the calculation of the representative views set. In the exploration of medical datasets, it is difficult to simultaneously visualize interior and exterior structures because the structures are commonly quite complex and it is easy to lose the context. We have developed a new interaction tool, the Virtual Magic Lantern, tailored to facilitate volumetric data inspection in a Virtual Reality environment. It behaves like a lantern whose illumination cone determines the region of interest. It addresses the occlusion management problem and facilitates the inspection of inner structures without the total elimination of the exterior structures, offering in this way, a focus+context-based visualization of the overall structures. Finally, the analysis of medical datasets may require the selection of 3D points for measurements involving anatomical structures. Although there are well-established 3D object selection techniques for polygonal models, there is a lack of techniques specifically developed for volume datasets. We present a new selection technique for Virtual Reality setups which allows users to easily select anchor points in non-necessarily segmented volume datasets rendered using Direct Volume Rendering. This new metaphor is based on the use of a ray emanating from the user, whose trajectory is enriched with its points of intersection with the on-the-fly determination of the isosurfaces along the ray path. Additionally, a visual feedback of the ray selection is offered through the use of two helper mirror views, in order to show occluded candidate points that would otherwise be invisible to the user without posterior and ad-hoc manipulation.
