OA136] The lumic challenge: How accurate are registration algorithms in registering lung CT images for subtraction?

Abstract

Purpose To determine the accuracy of registration algorithms, using an anthropomorphic digital phantom, in registering unenhanced and enhanced computed tomography (CT) chest images. Methods Lung subtraction CT generates pulmonary iodine maps by subtracting an unenhanced scan from an enhanced scan after image registration. This technique can be useful for detection of pulmonary embolism, characterization of nodule or mass perfusion, or for evaluation of pulmonary lesion growth at follow-up. The LUMIC challenge is aimed at comparing the performance of algorithms in registering simulated enhanced and unenhanced CT images of the XCAT anthropomorphic digital phantom (Duke University, Durham, North Carolina, U.S.A.). The simulated scans are available with two voxel sizes (anisotropic, 0.6 × 0.6 × 1.0 mm3, and isotropic, 1.0 × 1.0 × 1.0 mm3), and with varying differences in diaphragm levels between the CT scans: 3 mm (small difference), 8 mm (average clinical difference), and 20 mm (large difference), all without pulmonary pathology. Since the voxel-by-voxel true displacement between the unenhanced and enhanced phantom realizations is known, the residual error between the estimated displacement field and the known truth can be determined for all lung voxels. Results The results submitted in response to this challenge will be calculated as the median, 25th and 75th percentile of the residual values. Besides these values, whole residual error images will be generated, which will provide position-specific information on the performance of each algorithm. Three algorithms have already been evaluated, resulting in the following residual errors for the large diaphragm-position difference [median (25th – 75th percentile)]: 0.93 mm (0.52 mm – 1.58 mm), 0.92 mm (0.51 mm – 1.57 mm), and 0.93 mm (0.52 mm – 1.60 mm). For all three algorithms, the largest errors were seen in the paracardiac regions and close to the diaphragm. Conclusions This challenge, based on the use of a digital anthropomorphic phantom, is a useful way to not only determine the current performance level of registration algorithms world-wide, but also to directly compare their performance. In addition, given the availability of the resulting error location information, the participants will be able to identify problematic areas for their registration algorithms that might need further improvement. The challenge is available on https://lumic.grand-challenge.org .