An articulated shape model to predict paediatric lower limb bone geometry using sparse landmarks

Abstract

Creating musculoskeletal models in a paediatric population currently involves either creating an image-based model from medical imaging data or a generic model using linear scaling. Image-based models provide a high level of accuracy but are time-consuming and costly to implement, on the other hand, linear scaling of an adult template musculoskeletal model is faster and common practice, but the output errors are significantly higher. An articulated shape model incorporates pose and shape to predict geometry for use in musculoskeletal models based on existing information from a population to provide both a fast and accurate method. From a population of 333 children aged 4–18 years old, we have developed an articulated shape model of paediatric lower limb bones to predict bone geometry from eight bone landmarks commonly used for motion capture. Bone surface root mean squared errors were found to be 2.63 ± 0.90 mm, 1.97 ± 0.61 mm, and 1.72 ± 0.51 mm for the pelvis, femur, and tibia/fibula, respectively. Linear scaling produced bone surface errors of 4.79 ± 1.39 mm, 4.38 ± 0.72 mm, and 4.39 ± 0.86 mm for the pelvis, femur, and tibia/fibula, respectively. Clinical bone measurement errors were low across all bones predicted using the articulated shape model, which outperformed linear scaling for all measurements. However, the model failed to accurately capture torsional measures (femoral anteversion and tibial torsion). Overall, the articulated shape model was shown to be a fast and accurate method to predict lower limb bone geometry in a paediatric population, superior to linear scaling.