Accurate automated quantitative estimation of spinal curvature is an important task for the clinical evaluation and treatment planning of scoliosis.It solves the disadvantage of manual Cobb angle measurement(time-consuming and unreliable)which is the current clinical standard for scoliosis assessment.A couple of attempts have been made for automated Cobb angle estimation on single-view x-rays.However,it is very challenging to achieve a highly accurate automated estimation of Cobb angles because it is difficult to utilize the information of Anterior-posterior(AP)and Lateral(LAT)view x-rays efficiently.We therefore propose a novel Multi-View Extrapolation Net(MVE-Net)architecture that can provide an accurate automated framework for scoliosis estimation and error correction in multi view(both AP and LAT)x-rays.The proposed MVE-Net consists of three closely-linked components:(1)a joint-view net learning AP and LAT angles jointly based on landmarks learned from joint representation,(2)an independent-view net learning AP and LAT angles independently based on landmarks learned from unique independent feature of AP or LAT angles,and(3)an inter-error correction net learning a combination function adaptively to offset the errors of the first two nets for accurate angle estimation.Experimental results on 526 x-ray images show an impressive 7.81 Circular Mean Absolute Error(CMAE)in AP Cobb angle and 6.26 CMAE in LAT Cobb angle estimation,which demonstrates the MVE-Net's capability of performing accurate estimation of Cobb angles in multi-view x-rays.Our method therefore provides clinicians with a framework for efficient,accurate,and reliable estimation of spinal curvature for comprehensive scoliosis assessment. |