Reconstructing bones: using statistical shape modelling to create 3D models of the femur from ultrasound images

Abstract

Many children with cerebral palsy (CP) develop bony deformities of the femur that require surgical intervention to correct1. Concerns regarding the radiation exposure from CT and the cost and scan time of MRI mean patient-specific 3D models of the femur are rarely used for surgical planning in this patient group, despite evidence supporting their role in improving surgical outcomes2,3,4. Ultrasound (US) imaging presents a cheap, low-risk, and readily available means of constructing such models. However, US is only able to capture partial views of the femur. The “missing” views may be reconstructed using statistical shape modelling; a mathematical technique used to quantitatively analyse complex shapes5,6. Can patient-specific 3D models of the femur be accurately reconstructed from partial surface data acquired with simulated 3D ultrasound using statistical shape modelling? 60 3D meshes of the femur were derived from MR images of 32 young adult subjects (13 with CP, 19 typically developing (TD)). The femur meshes from the left side were flipped horizontally to match those from the right. The meshes from both groups were then used to construct a statistical shape model (SSM) of the femur. An algorithm was written which used the SSM to reconstruct a complete femur mesh from partial information. To test the effectiveness of the algorithm, a dataset of partial surfaces replicating the views possible using US was created. Complete femurs were reconstructed from this dataset, and evaluated against the original 3D meshes using a leave-one-out cross validation procedure. An average point-to-point error of 1.16 ± 0.45 mm was found for reconstructions of the femurs from the TD group, compared to 2.55 ± 0.47 mm in the CP group. Fig. 1 – “a) Example partial surface from the simulated US dataset; b) Example TD reconstruction; c) Example CP reconstruction (reconstruction in purple, original mesh in white”)Download : Download high-res image (36KB)Download : Download full-size image The relatively low error for the reconstructions of the TD femurs demonstrates a promising proof of concept for the proposed technique of creating 3D femur models from partial surface data acquired with US. Future work may develop the algorithm further to improve its performance in the presence of increased femoral deformity, as found in the CP group. With development, this technique has the potential to bring the use of 3D models for preoperative planning into common practice for this patient group, which is likely to improve surgical outcomes. Although the focus of this study has been the creation of 3D models of the femur, the technique of reconstructing US images using statistical shape modelling could be applied to other anatomical structures. Owing to the reduced risk, cost and scan time compared with CT and MRI, the application of the proposed reconstruction technique has the potential to positively impact other surgical services.