Simulation and Interaction Methods for Intuitive Deformation of Volume Images
The areas of simulation and visualization in computer graphics offer technical solutions that can aid medical applications in multiple ways, such as visualization and exploration of volume images to detect tumors or other diseases, or simulate the deformation of organs to train surgeons or plan surgical interventions. However, enabling the simulation of deformations on a medical image of a patient requires many pre-processing operations, which are time-consuming. These operations include image segmentation, meshing of anatomical elements, or assignment of mechanical parameters and boundary conditions. Depending on the organ to be deformed, different parameters and settings must be specified, which requires additional workload and a per-case adaptation of the simulations and the pre-processing. This thesis starts from the grand goal of empowering radiologists and surgeons with a 3D environment where they can easily and intuitively interact with a patient’s volume image. We stand on important conditions, such as the lack of knowledge of simulation methods by clinical experts, and the need to create tools with a simple learning curve, and without pre-processing needs. For instance, surgeons should be able to load the medical image of a patient, and directly interact with the organs present in this image, without complex segmentation or case-dependent parameter setting. The simulation tool should allow surgeons to move organs using a computer mouse, or even their fingers directly on a touchscreen. By avoiding tedious and complex processes such as segmentation, we can reduce the cost and complexity to use simulation tools, and increase their usability for more patient cases. Rapid setup of medical simulation is a radically new goal with respect to most previous research. Standing on the aforementioned goals and conditions, we have developed novel deformable models that allow simple and interactive manipulation of volumetric medical images, and can therefore enable the creation of simple tools for exploration and planning of surgical interventions. Our novel models and methods can be classified as follows: First, we have designed an interactive algorithm to deform 3D images without segmentation through a corotational coarsening method, which uses a coarse and a fine mesh. The fine mesh captures the properties of the volume at a fine level, and the coarse mesh allows very fast simulations. Each coarse tetrahedron represents the material of the under lying fine tetrahedra. The meshes can be either regular or irregular, although we demonstrate the method on regular meshes for simplicity. The coarsening method enables the simulation of the tetrahedral mesh, and then we apply a rasterization method to deform the full volume from the coarse nodes. Second, we add accurate boundary conditions at arbitrary locations within the volume. The previous step assumes the application of forces at coarse nodes, but we desing additional methods that demonstrate how to apply external forces at fine nodes without the need to resolve the simulation system at a fine level. We take into account fixed and moving fine nodes inside each coarse tetrahedron, and we develop a corotational coarsening formulation where fixed fine nodes affect directly the behaviour of coarse nodes. Finally, we create an interaction method that allows a person to move 3D organs using fingers on a 2D touchscreen. Motion is imposed using the fingers, which gives the user the sensation of grasping and rotating anatomy directly. We demonstrate this novel interaction method on a volume image that contains the back bone, kidneys, veins, the stomach and other soft anatomy in the abdomen region. The sensation of direct manipulation is provided by maintaining the same anatomical elements always under the fingers, as the interaction evolves. In conclusion, we have developed a variety of simulation and visualization methods that allow non-technical users to deform and manipulate anatomical elements in medical volume images in an intuitive way, directly on a touchscreen.
Tesis Doctoral leída en la Universidad Rey Juan Carlos de Madrid en 2019. Director de la Tesis: Miguel Ángel Otaduy
- IA - Tesis Doctorales