Research topic: Medical Simulation
Projet: Mimesis http://mimesis.inria.fr/
Scientifique Responsable: Stéphane Cotin
“Everything should be made as simple as possible, but not simpler”
This quote describes succinctly the main constraint in the development of real time simulations of soft tissues, model complexity and overly refined meshes should be sacrificed in order to maximize the frame rate. Indeed, real-time computation is a major requirement for simulations in the context of both training and assistance in the operating room . Yet, providing solutions for accurate, patient-specific biomechanical modeling is essential. One of the main research interests of the MIMESIS team is to propose numerical methods and dedicated models to meet real-time constraints (for instance, the use of Euler Bernoulli beam theory over 3D elasticity ). Despite these models are most of the time validated experimentally in static cases , there is a need to provide feedback to the surgeons allowing for the estimation of the error and the reliability of the models during the simulation.
In this position, the candidate will develop error estimates for two types of errors. First, we will investigate the discretization error which assesses the validity of the mesh. Second, we will study the model error which characterizes the validity of the surrogate/simplified models used in the simulation . In this context surrogate models mean, for instance, the use of beam theory instead of 3D elasticity or the use homogenised models instead of the real multiphase materials. Besides the development of such estimates, the candidate will also develop error indicators and strategies for adaptivity. The error indicators will enable us to identify of the regions that contribute most to the error whilst the adaptive strategies will enable us to apply corrective measures when required, i.e. mesh refinement for high discretization error or reintroduction of a more complex description for high model error. As a result, this will allow us to keep the accuracy under control while at the same time limiting the computational cost. This idea has already been tested and a first journal article on the topic of error-controlled refinement is under revision. As done for this preliminary work, the different developments will be implemented in the open source framework SOFA (www.sofa-framework.org).