Investigation of the Sensitivity of Ventilation Accuracy to the Deformable Image Registration Regularization Parameter
H Byrne1,3*, J Kipritidis2, O Dillon1, J Booth2,3, P Keall1, (1) ACRF ImageX Institute, Faculty of Medicine and Health, The University of Sydney, Australia, (2) Northern Sydney Cancer Centre, Royal North Shore Hospital, Australia, (3) Institute of Medical Physics, School of Physics, University of Sydney, Australia
Presentations
MO-IePD-TRACK 4-3 (Monday, 7/26/2021) 5:30 PM - 6:00 PM [Eastern Time (GMT-4)]
Purpose: To investigate the sensitivity of producing accurate ventilation information from deformable image registration (DIR) to the regularization parameter, λ, by comparing DIR-derived ventilation with Galligas PET images for a range of values of λ.
Methods: Paired inhale/exhale breath hold CT scans and Galligas PET ventilation images were obtained from 8 patients. The exhale CT was deformably registered to the inhale CT for a range of λ from 0.05 to 1.5. Two previously described methods were used to produce ventilation maps: first the Hounsfield unit change between the deformed inhale and exhale image to infer lung density change, second the Jacobian determinant of the deformation vector field to infer air volume change. Three assessment metrics were used (1) Spearman correlation between the CT ventilation maps and the Galligas PET images, (2) percentage of negative values in the Jacobian determinant (indicating singularities in the motion field, see AAPM report TG132) and (3) mean square error between the exhale and deformed inhale CT images.
Results: Spearman correlations between CT ventilation maps and the Galligas PET images generally increased as λ increased from 0.05 to around 0.75, with the gains stabilizing as λ was increased further. Image match accuracy measured by mean square error increases with smaller λ as expected. However, λ < 0.5 created motion models implying up to 6.3% of the lung contracted during inhale, and derived ventilation maps using the Jacobian determinant had Spearman correlations as low as 0.20, implying non-physiological deformation. The dependence of the Spearman correlation on λ was more marked for the Jacobian determinant method than the Hounsfield unit method.
Conclusion: Physiological ventilation information derived from deformable image registration using two ventilation algorithms was benchmarked against Galligas PET as the regularization parameter was varied. The Hounsfield unit algorithm proved robust across the range of λ tested.
Funding Support, Disclosures, and Conflict of Interest: This work was supported by two government grants: a Cancer Institute NSW Translational Program Grant and an NHMRC Investigator Grant. Paul Keall is an inventor of a US patent on CT ventilation (#7668357). This patent is owned by Stanford University and is unlicensed.
ePosters
Keywords
Functional Imaging, Deformation, CT
Taxonomy
IM- CT: Quantitative imaging/analysis
Contact Email