2D/3D Reconstruction of The Distal Tibiofibular Joint from Biplanar Radiographs Using Deep Learning Registration and Statistical Shape and Intensity Model

In the present study, we introduce a novel approach for the three-dimensional (3D) reconstruction of the distal tibiofibular joint shape from its two-dimensional (2D) biplanar radiographs. An independent Statistical Shape and Intensity Model is used to generate radiographs using the mean model and its associated variations. These generated Anteroposterior and Lateral images are subsequently utilized to train a Deep Learning-based regressor network, allowing for the instantaneous determination of the joint's shape, intensity, and pose parameters in an end-to-end fashion. This leads to expeditious reconstruction of the 3D surface of the anatomical tibiofibular joint's 3D surface from its 2D radiographs. Our methodology was applied to reconstruct the distal tibia and distal fibula simultaneously. The results demonstrate that the Deep Network Regressor is proficient in reconstructing the surface of the entire structure with an average error of 0.63 millimeters.