A systematic approach to feature tracking of lumbar spine vertebrae from fluoroscopic images using complex-valued wavelets.

This paper presents a systematic approach to lumbar spine vertebrae tracking in fluoroscopic images using complex-valued wavelets. The proposed algorithm is designed specifically based on a set of performance criteria associated with the detection and tracking of feature points in lumbar spine vertebrae from fluoroscopic images. The algorithm handles contrast and illumination non-homogeneities and noise in fluoroscopic images through the use of local phase information obtained using complex-valued wavelets. The algorithm is capable of tracking feature points that undergo various geometric deformations caused during the fluoroscopic imaging process by defining a descriptor that is invariant to scale and rotation and robust to affine, projective and mild pin-cushion distortions. The algorithm has been tested using dynamic sagittal fluoroscopic videos of the lumbar-sacral region and testing results indicate that the algorithm achieves good tracking performance of lumbar spine vertebrae in fluoroscopic images that exhibit contrast and illumination non-homogeneities as well as noise, with mean root mean square error of less than 0.40 mm under in all test sequences.