Revisiting Contour-Driven and Knowledge-Based Deformable Models: Application to 2D-3D Proximal Femur Reconstruction from X-ray Images

In many clinical applications, 3D reconstruction of patient-specific structures is major interest. Despite great effort put in 2D-3D reconstruction, gold standard bone obtained by segmentation on CT images still mostly used – at the expense exposing patients to significant ionizing radiation and increased health costs. State-of-the-art methods are based non-rigid registration digitally reconstructed radiographs (DRR) aiming full automation but with varying accuracy often exceeding requirements. Conversely, contour-based approaches can lead accurate results strongly depend quality extracted contours have been left aside recent years. this study, we revisit a method for proximal femur contours, image cues, knowledge-based deformable models. statistical shape models were built using 199 scans from THA that generate pairs high fidelity DRRs. Convolutional neural networks trained DRRs investigate automatic contouring. Experiments conducted DRRs, calibrated pelvis phantom volunteers an analysis contouring its automatization. Using manual DRR, best error was 1.02 mm. With state-of-the-art femur, highlighted relevance challenges contour-driven yield