Real-time Biomechanical Modeling for Intraoperative Soft Tissue Registration

Stefan Suwelack
2015
Computer assisted surgery (CAS) systems intraoperatively support the surgeon by providing information on the location of hidden risk (e.g. vessels, nerves) and target structures (e.g. tumors) during surgery. In this way CAS techniques have been a major driving force for improving patient outcomes for many applications such as orthopedic surgery and neurosurgery. However, CAS is currently not used in the daily clinical routine for laparoscopic interventions. The main reason for this discrepancy
more » ... re soft tissue deformations that make intraoperative registration (and thus intraoperative navigation) difficult. In this work, a novel, biomechanics based approach for real-time soft tissue registration from sparse intraoperative sensor data such as stereo endoscopic images was developed. At the core of the method lies an accurate, yet real-time capable finite element (FE) model of the liver. For this purpose, a novel GPU based multigrid finite element solver is presented. The solver is based on a novel mapping scheme that allows to transfer displacements and forces between unstructured, non-conforming, higher order meshes. In order to achieve high efficiency on parallel hardware, a sparse approximate inverse approach is used for preconditioning and smoothing. By pre-computing and subsequently adapting this operator to the current deformation each time step, the approach becomes real-time capable. In order to match the preoperative organ model to an intraoperative partial surface, the novel Physics based Shape Matching (PBSM) scheme is presented. This approach treats the non-rigid surface registration as an electrostatic-elastic problem, where an elastic body that is electrically charged (preoperative model) slides into an oppositely charged rigid shape (intraoperative surface). In contrast to previous attempts at biomechanically based registration, this novel physics based interpretation allows casting the shape matching problem into a single variational formulation. It is also the first method that employs a non-linear, yet realtime capable biomechanical model of the liver for registration purposes. In a large validation study based on numerical and phantom data, it was shown that the novel method outperforms state-of-the-art algorithms. Further contributions of this work include methods for simulating tissue cutting during the intervention as well as new validation tools for CAS systems. To my wife Eva-Maria
doi:10.5445/ksp/1000046886 fatcat:l7lu2y5635gvrkmqn5dp3fytum