The motivation behind this project is the need for 3D representation by doctors in Orthodontics to assist diagnosis. However the limit on the amount of dose could be used for pediatric patients inhibit the amount of CT scans could be done. There are other systems on the market such EOS Imaging Flex Dose which is based on an atlas based registration method to reconstruct two 2D x-ray projections into 3D spinal representation. However such system is not only very expensive (selling in USA at $1M for one system) but also produce certain image artifacts due to the atlas is based on normal people. This work is to show the feasibility of deep learning could surpass performance of these other systems and achieve high resolution spinal 3D representations using 2D images.