Patient-Specific 3D Reconstruction of a Complete Lower Extremity from 2D X-rays

Introduction In clinical routine surgeons depend largely on 2D x-ray radiographs and their experience to plan and evaluate surgical interventions around the knee joint. Numerous studies have shown that pure 2D x-ray radiography based measurements are not accurate due to the error in determining accurate radiography magnification and the projection characteristics of 2D radiographs. Using 2D x-ray radiographs to plan 3D knee joint surgery may lead to component misalignment in Total Knee Arthroplasty (TKA) or to over- or under-correction of the mechanical axis in Lower Extremity Osteotomy (LEO). Recently we developed a personalized X-ray reconstruction-based planning and post-operative treatment evaluation system called “iLeg” for TKA or LEO. Based on a patented X-ray image calibration cage and a unique 2D–3D reconstruction technique, iLeg can generate accurate patient-specific 3D models of a complete lower extremity from two standing X-rays for true 3D planning and evaluation of surgical interventions at the knee joint. The goal of this study is to validate the accuracy of this newly developed system using digitally reconstructed radiographs (DRRs) generated from CT data of cadavers. Methods CT data of 12 cadavers (24 legs) were used in the study. For each leg, two DRRs, one from the antero-posterior (AP) direction and the other from the later-medial (LM) direction, were generated following clinical requirements and used as the input to the iLeg software. The 2D–3D reconstruction was then done by non-rigidly matching statistical shape models (SSMs) of both femur and tibia to the DRRs (seee Fig. 1). In order to evaluate the 2D–3D reconstruction accuracy, we conducted a semi-automatic segmentation of all CT data using the commercial software Amira (FEI Corporate, Oregon, USA). The reconstructed surface models of each leg were then compared with the surface models segmented from the associated CT data. Since the DRRs were generated from the associated CT data, the surface models were reconstructed in the local coordinate system of the CT data. Thus, we can directly compare the reconstructed surface models with the surface models segmented from the associated CT data, which we took as the ground truth. Again, we used the software Amira to compute distances from each vertex on the reconstructed surface models to the associated ground truth models. Results When the reconstructed models were compared with the surface models segmented from the associated CT data, a mean reconstruction accuracy of 1.2±0.2mm, 1.3±0.2mm, 1.4±0.3mm and 1.3±0.2mm was found for left femur, right femur, left tibia and right tibia, respectively. When looking into the reconstruction of each subject, we found an average reconstruction accuracy in the range of 1.1mm to 1.5mm. Overall, the reconstruction accuracy was found to be 1.3±0.2mm. Discussions We presented a cadaver study to validate the accuracy of reconstructing 3D patient-specific models of a complete lower extremity from 2D X-rays. Our experimental results demonstrate that the complete lower extremity can be reconstructed accurately from 2D X-rays. Please note that the errors we reported above include both pose and shape reconstruction errors whole most of previous studies only reported shape reconstruction errors.