Predicting tibia and fibula geometry from shank surface skin in a paediatric population

Musculoskeletal model generation often requires scaling a generic adult skeleton, which is fast but generally inaccurate. Statistical shape models have been developed to scale musculoskeletal models based on the population of interest, improving personalisation. Low-cost 3D surface scanning presents a new opportunity to scale musculoskeletal models directly from external body shape, such as SKEL. This study assessed the accuracy of 1) a combined shape model to predict tibia/fibula geometry from skin surface in a typically developed paediatric population and 2) the output from the generic scaling through SKEL. The left tibia/fibula bones and shank skin surface were segmented from 318 full-body CT scans. Principal component analysis captured the combined morphological variations between the shank surface and tibia/fibula geometry. A combined shape model was then developed to predict tibia/fibula geometry from the external shank surface using Principal Component fitting. The model was validated using a leave-one-out analysis. The combined shape model predicted the tibia and fibula with a root-mean-square error of 2.00 ± 0.90 mm, whereas the surface error from SKEL fitting was 2.98 ± 0.62 mm. Clinical bone measurement derived from predicted bones showed low error when compared to the segmented bones, except for tibial length and the mechanical medial proximal tibial angle. These results were repeatable across three prediction rounds, demonstrating robust and accurate prediction of bone shape and clinical measurements from the external skin surface. This proof-of-concept demonstrates the feasibility of inferring bones from the skin, paving the way for full-body models to enhance paediatric clinical gait analysis.