Fast quasi-automated 3D reconstruction of lower limbs from low dose biplanar radiographs using statistical shape models and contour matching.

Three-dimensional bone reconstructions from medical imaging are essential for biomechanical modelling and are growing tools in clinics. Several methods of lower limbs reconstruction from biplanar radiographs have been proposed in the literature but with significant operator dependence. A novel reconstruction method based on reduced manual annotation, statistical shape models and fully automatic adjustments was proposed in this study. While significantly reducing operator intervention, the proposed method demonstrated similar or better precision than previous approaches on clinical parameters. Meanwhile, shape accuracy was improved to around 1mm. By quasi-automating the 3D reconstruction without loss of accuracy and precision, the proposed approach is a considerable step towards extensive use of 3D personalized models in clinical routine and large cohort biomechanical studies.