Program Information
Shape Attribute Registration for Matching of Pulmonary Vasculature in Thoracic Image Registration
G Hugo1*, E Weiss1 , C Guy1 , M Riblett1 , Y Pan2 , G Christensen2 , (1) Virginia Commonwealth University, Richmond, VA, (2) University of Iowa, Iowa City, IA
Presentations
MO-F-205-7 (Monday, July 31, 2017) 4:30 PM - 6:00 PM Room: 205
Purpose: To develop a shape attribute method for image registration to manage large anatomical changes observed during radiation therapy for lung cancer.
Methods: We extend the concept of using moments of 1D image intensity histograms to improve the specificity of registration and avoid mismatch. The vessel information in a CT image is extracted using a vesselness filter and a 2D histogram (intensity, spatial radius) is computed at each voxel, at three scales. Nine geometric moments at each scale are computed and collected into a shape attribute vector. These vectors are matched between images using a sum of square difference (SSD) cost. This sum of square attribute difference (SSAD) registration was compared to conventional SSD and mutual information (MI) registrations in clinical images. Artificial deformations were applied to 2D retinal images to generate 100 sample image pairs for sensitivity analysis, as 2D images are easy to interpret. Sensitivity and accuracy of the SSAD registration were evaluated as a function of histogram parameters. Vesselness images from CTs acquired during radiation therapy, before and after atelectasis resolution, in 5 patients were registered and accuracy assessed by landmark error.
Results: In 2D tests, best performance was achieved with a radius of 5 voxels, 5 or 10 radial and intensity bins, and keeping the radius fixed across three resolution levels. For large deformations (maximum displacement greater than 50 voxels), mean accuracy was 8.1, 11.3, 12.9% of applied deformation for SSAD, SSD, MI. Accuracy was significantly higher for SSAD compared to SSD (p<1e-05, signed rank test) and MI (p<1e-05). In clinical images, landmark error was 4.2 ± 6.0 mm for SSAD and 4.6 ± 6.7 mm for SSD.
Conclusion: Shape attribute matching improved performance over conventional algorithms in matching lung anatomy under large deformation, and may be useful to support adaptive radiotherapy.
Funding Support, Disclosures, and Conflict of Interest: This study was supported by a research grant from the NIH under Award No. R01CA166119. GDH and EW receive research support from Philips Healthcare and Varian Medical Systems and have licensed technology to Varian Medical Systems. EW receives royalties from UpToDate. GEC has licensed technology to Vida Diagnostics.
Contact Email: