Deformable Registration Techniques for Dual Energy Imaging in Clinical Radiotherapy
M Hoggarth*, J Luce, T Bray, S Nagda, J Roeske, Loyola Univ Medical Center, Maywood, ILTH-E-218-2 Thursday 1:00:00 PM - 2:50:00 PM Room: 218
Purpose: To demonstrate a deformable registration algorithm to reduce cardiac and respiratory motion artifacts in planar dual energy (DE) subtraction images obtained using a clinical on-board imaging (OBI) system.
Methods: A Demons-based deformable registration algorithm was developed to reduce motion artifacts between sequential planar x-ray images obtained on the OBI. The algorithm was applied to paired (120 kVp and 60 kVp) chest x-rays, deforming the low energy image to the high energy image. To test the algorithm, a total of 20-paired scans were obtained from 6 lung cancer patients. All images were obtained using respiratory gating (RPM, Varian Medical Systems, Palo Alto, CA). In order to quantify the reduction in image artifacts, homologous landmarks were chosen using soft-tissue features on both sets of images. Landmarks were placed throughout the images to examine both cardiac and respiratory motion. The root-mean-square (RMS) difference between the individual points was calculated both before and after application of deformable registration.
Results: A total of 160 landmarks were evaluated. Measurement of the RMS distances between the landmarks in the high and low energy deformed images showed improvement in all sectors. For points with < 2 mm RMS difference in original image, the deformation algorithm reduced the RMS residual by 36%. For points with > 2 mm RMS difference, the corresponding reduction was 64%. Qualitatively, DE images produced using Demons deformation showed fewer artifacts, and more defined tumor edges.
Conclusions: The Demons algorithm reduced motion artifacts across all DE planar images. The greatest improvements were observed in regions near the diaphragm and the heart. In particular, artifacts due to cardiac motion showed the most improvement as these are not taken into account when performing respiratory gating. The iterative, matrix-based algorithm would greatly benefit from integration of GPU parallel-processing and will be implemented in future investigations.