Estimation of Large Motion in Lung CT by Integrating Regularized Keypoint Correspondences into Dense Deformable Registration
J. Ruhaak, T. Polzin, S. Heldmann, I. Simpson, H. Handels, J. Modersitzki, and M. Heinrich
IEEE Trans Med Imaging, vol. 36, issue 8, August 2017
The accurate registration of inspirative and expirative pulmonary CT scans has numerous applications in the diagnosis and analysis of respiratory diseases such as chronic obstructive pulmonary disease (COPD). For a reliable diagnostic use, the computed deformation fields must align corresponding structures correctly, while at the same time estimate the occurring local volume change during respiration in a physiologically plausible way. Satisfying these partially competing objectives has so far remained an unsolved problem.
We present a deformable image registration algorithm that integrates correspondence information obtained by regularized keypoint matching (see left images) into a dense deformable registration framework (right images). The discrete matching jointly considers several thousands of displacement vectors for a large number of interest points. The main registration yields a realistic estimation of the local volume change by restricting the Jacobian determinant of the transformation. An average runtime of five minutes permits clinical adoption.
The proposed algorithm ranks first in the EMPIRE10 challenge on pulmonary image registration. Moreover, it achieves an average landmark distance of 0.82 mm on the DIR-Lab COPD database, thereby improving upon the state of the art in accuracy by 15 %. Our algorithm is the first to reach the inter-observer variability in landmark annotation on this challenging dataset.