high performance computing on graphics processing units: hgpu.org

Many problems can occur during childbirth which may lead to instant or future morbidity and even mortality. Therefore the computer-based simulation of the mechanisms and biomechanics of human childbirth is becoming an increasingly important area of study, to avoid potential trauma to the baby and the mother throughout, and immediately following, the childbirth process. Computer-based numerical methods, such as the Finite Element Method, have become more widespread to simulate biological phenomena over the last two decades or so. One of the important aspects of such methods is them being able to accurately model the underlying physics and biomechanics of biological processes. In the case of the childbirth process, an important role is played by the fetal head and its motion as it is being born. The most important manifestations of the head’s motion are described as the cardinal movements. Being able to model the cardinal movements in a computer-based model of the human childbirth process is compulsory as they occur in almost every normal delivery. Many existing simulations use reverse-engineered approaches to model the cardinal movements by imposing pre-defined trajectories that approximate a real childbirth. These approaches lack physical accuracy and are unable to extend the simulation to unseen scenarios where for example the childbirth process does not develop normally. To create a simulation software capable of simulating realistic, including unseen, scenarios, and which does not make any pre-simulation assumptions about the cardinal movements, a physical and forward-engineered approach in which the motions of the head are determined by the underlying physics, is required. This thesis presents a simulation system where the physical behaviour of the fetal head is modelled during the second stage of childbirth. Accurate models of the fetal head and trunk, the maternal uterine cervix, bony pelvis and pelvic floor muscles, were created. Some of these are considered to be rigid bodies in the simulation (fetal head, trunk and maternal bony pelvis), whereas others are considered to be deformable (maternal uterine cervix and pelvic floor muscles. The Finite Element Method (FEM) is used to model the deformable tissues by implementing the Total Lagrangian Explicit Dynamics (TLED) approach on the GPU. The combined TLED/contact mechanics method was first tested on simple hyperelastic objects. Following successful validation, the interaction between the fetal head and the deformable tissues was evaluated using a projection based contact method in conjunction with TLED. Several experiments had to be done to find the required set of factors contributing to the occurrence of the cardinal movements. These steps are reported in the thesis as well as the results of the final experiments which do exhibit the key cardinal movements of a normal childbirth process, marking the successful, and key, contribution of this thesis. The GPU based acceleration allows running the simulation in near real-time, which makes it possible to create interactive simulations for training purposes of trainee obstetricians and midwives. The simulation system presented in this work is also the first step towards a fully patient specific system that would allow clinicians to diagnose and/or predict adverse scenarios of childbirth based on the MRI scans of real pregnancies prior to labour.