Assessment of hip deformities during childhood: a combined approach of Machine Learning and Finite Element Analysis

Abstract: Osteoarthritis is a degenerative disease that will cause pain and stiffness in the joint if not treated. One cause of early osteoarthritis in the hip joint is pediatric hip deformities. Hip deformities can occur in different stages of childhood and cause the bones to be misshapen or the hip joint to be dislocated. This thesis aimed to build subject-specific computer models of the hip joint to find a biomechanical connection between pediatric hip deformities and osteoarthritis. The aim was approached through three main tasks: - Automatically detect anatomical landmarks on two-dimensional X-ray images. This can be used to automate three-dimensional reconstructions of the femur. - Predict the effects of common loading scenarios in the hip joint cartilage using subject-specific numerical models developed using X-ray Computed Tomography (CT) images. The model can later be customized to fit a reconstructed femur. - Assess the feasibility of obtaining a three-dimensional subject-specific numerical model of the hip joint from a two-dimensional X-ray image to predict the mechanical effect of common loading scenarios in the cartilage. For the methodology, Dual-energy X-ray Absorptiometry images of hip joints were used to evaluate and refine a machine learning based method for automatic landmark detection. Additionally, a finite element model was developed based on a CT image and it modeled one-leg-stance and walking. Finally, three-dimensional reconstructions of the femur were made based on automatically detected landmarks on two-dimensional X-ray images and used to scale the finite element model to match the size of each reconstructed femur. Results showed that the method for automatic landmark detection was able to identify eight anatomical landmarks in the hip joint with an average error of 0.33 mm when compared to manually annotated landmarks. The finite element model of the hip joint agreed well with literature when calculating the contact pressure in the cartilage during standing and walking. Finite element models scaled in three directions based on reconstructed femurs were able to predict hip joint cartilage mechanics. As this project showed that the concept of creating a three-dimensional finite element model based on two-dimensional X-ray images was successful, future research should focus on automating the process further and making it applicable to children to be more clinically relevant. This could enable the possibility to gain information on how pediatric hip deformities affects the cartilage in the hip joint without additional radiation exposure.