In the biomedical field, risk models, which provide a probability of occurrence of a given pathology, are typically built using clinical data. The predictive ability of risk models such as logistic regression or support vector machines is therefore highly dependent on the amount of available information and the problem dimensionality. At the other end of the spectrum, fields such as engineering tend to rely increasingly on the predictions of computational models (e.g., finite element simulations, CFD). Despite their vast differences, both approaches can in fact be combined to increase the predictive ability of risk models. This can be achieved, for example, by using the outcome of numerical simulations to complement the clinical data.
This seminar will present several approaches to perform such a fusion of
data in the case of hip fracture risk prediction. Computational data from finite
element simulations will be used to generate complementary information to refine the fracture risk model. This information will also provide a mechanical insight typically missing from traditional risk models.
The presentation will first provide the necessary background in risk model
construction and hip fracture. The approaches for data fusion will then be
described and demonstrated on an actual clinical dataset.