Shape and pose estimation of a skeleton in X-ray images using statistical shape and intensity model

Abstract

Bi-planar fluoroscopic (BPF) image analysis is one of the methods for accurately measuring the joint’s position and posture and allows the study of the joint kinematics in-vivo without being invasive. In this study, a method to find the shape and position of bones using the X-ray images’ brightness was proposed. Existing methods used CT or MRI and labeled it to obtain a volume image to get the bone’s actual shape, and the labeled bones were used for registration. However, this study presents a method of predicting shape and posture based on the shape database without using bone models from volume images. A methodology for generating a statistical shape and intensity model (SSIM) from volume images were presented, and SSIM was used as a database for the bony shape. In addition, a method of generating digital fire composition pictures (DRR) from SSIM was also developed for comparison with actual X-ray images. The developed DRR generation method takes about 20 ms to generate an image of 512 × 512 pixels from the proposed SSIM. By changing the shape parameters of the proposed SSIM, a subject-specific shape was found. The position and posture were estimated by comparing the DRR with the two-direction X-ray images taken from the subjects. When the shape and position are matched simultaneously, the axial registration error has an average of 1.54 mm and 2.98˚, and the Hausdorff shape error has an average of 5.92 mm. If the most accurate shape parameter is found and fixed, the axial registration error has an average of 0.36mm and 0.82˚. The proposed registration method can be used to measure intra-articular skeletal movements such as knee and ankle joints. It can be used for image-guided radiation therapy and surgery and is expected to study patients’ skeletal movement patterns and characteristics.