An accurate geometrical three-dimensional (3D) model of human bone is required in many medical procedures including Total Knee Arthroplasty (TKA) and computer-assisted surgical navigation. Segmentation of Computed Tomography (CT) datasets is commonly used to obtain such models. However, such a method is expensive and time consuming. We herein propose a novel method for patient specific bone model reconstruction using standard x-ray fluoroscopy, a cheaper and widely available imaging alternative.

Fluoroscopic images are taken at multiple arbitrary viewpoints to provide sufficient information for bone reconstruction. The viewpoints can be obtained by either rotating the imaging source and detector or the patient’s limb of interest. The bone’s pose within the radiological scene in each of the captured images can be estimated by tracking a set of metallic calibration markers within a calibration target, rigidly attached to the limb of interest. Having acquired the required calibration data, a complex iterative scheme is executed to optimize a statistical bone atlas of the bone of interest and the relative pose between the bone and the calibration target.

In order to verify our method, we performed a cadaveric study. A set of rigidly attached fiducial markers were attached to a cadaveric leg. The leg was imaged using x-ray fluoroscopy while being rotated axially to provide us with the images required for bone model reconstruction. Distal femur and proximal tibia bone models were reconstructed from the fluoroscopy images. Furthermore, the leg was CT-scanned and segmented to provide us with the ground-truth required for reconstruction accuracy assessment. Results show the adequacy of the proposed method for surgical applications.