Panoramic Radiography and Cone Beam Computed Tomography (CBCT) are the most used imaging techniques for implant treatment and oral surgery. However, it cannot be used frequently due to the high cost and radiation exposure of CBCT. Whereas Panoramic x-rays do not provide the high dimensional information necessary for planning surgery. Our research aims to produce an automated method for reconstructing a 3D dental structure from a single panoramic X-ray. This paper presents a robust preprocessing pipeline of CBCT data which are exploited to train a deep learning model for reconstructing the 3D dental structure from a single 2D panoramic X-ray image. It includes a method of generating a pair of synthetic panoramic X-rays and the corresponding flattened 3D volume as ground truth from CBCT data.