by Professor María-Angeles Pérez Ansón
Summary: Nowadays, new generation of predictive models are being developed to perform high level musculoskeletal systems simulations. One of the main disadvantages of computational models is their high computational cost. Different techniques have been proposed to reduce the simulation time. On the one hand, a numerical extrapolation technique was developed to run real-time simulations of bone remodeling predictions. On the other hand, neural networks was used as a predictive tool of certain musculoskeletal problems.
The process of strain-adaptive bone remodeling can be described mathematically and simulated in a computer model, integrated with the finite element method (FEM). A bone remodeling model implemented in a FE code can determine the long-term behavior of bone and the impact on bone biomechanics produced by the incorporation of prosthesis. Bone remodeling simulation normally starts assuming a uniform bone density distribution. As a consequence of the sequentially loads application, bone material properties changed, till a stable bone density distribution is predicted. As a consequence, we should run the FE solver for a large number of time increments to find out the final density distribution using a bone remodeling algorithm. Two vector extrapolation methods, reduce ranked extrapolation (RRE) and minimal polynomial extrapolation (MPE), were used to reduce the simulation time.
Neural Networks can be combined with intensive finite element simulations to develop predictive tools. Two examples will be presented. A coupled parametric finite element (FE) model and Artificial Neural Networks (ANN) were used to predict the fracture risk for healthy and osteoporotic patient conditions and under different boundary conditions. Additionally, a methodology was developed that combines computerized tomography images with bone remodeling simulations and ANN to determine the subject-specific forces from the computer tomography images.