Application of Lambda: Subvoxel Image Similarity in Complex Images
Presentations
SU-H300-IePD-F8-3 (Sunday, 7/10/2022) 3:00 PM - 3:30 PM [Eastern Time (GMT-4)]
Exhibit Hall | Forum 8
Purpose: To assess a sub-voxel similarity metric as applied to whole lung images with known deformations in a low dose regime.
Methods: 25 free-breathing helical CT image sets were taken of 4 patients, with 1 image selected per patient to be analyzed. Noise was injected into each image to simulate low-dose scanning, resulting in a set of noise-injected images corresponding to 40 (original), 30, and 15 mAs. The lung region was identified using the Pulmonary Toolkit’s lung segmentation algorithm. The original 40 mAs image served as the reference and the 30 and 15 mAs images served as the comparison. All images were identical except for noise. Each 30 and 15 mAs image had a differential deformation applied along the superior/inferior axis. The deformation magnitude ranged from 0 to 2 mm at the superior and inferior tips of the lung segmentation mask, respectively. Each reference-comparison pair was smoothed using an isotropic Gaussian filter using kernels of 0.5, 1.0, and 1.5 mm. The segmented regions of each pair was analyzed using λ, based on the dose-distribution metric γ, with input parameters of dHU=30 HU and DTA=0.5 mm. The resulting λ distributions were investigated to determine if the relationship between λ and λ-angle, θ -- the angle that lambda makes to the spatial axes -- and the deformation magnitudes were consistent with previous work.
Results: We found that choice of smoothing kernel heavily impacted the stability of the analysis, with the 1.0 and 1.5 mm kernels required to fully stabilize the θ distribution. A stabilized, |θ|≤30° restricted λ was able to qualitatively distinguish deformations as small as 0.5 mm.
Conclusion: λ was effective at detecting 0.5 mm deformations applied to whole lung images. Further analysis is needed to explicitly determine the accuracy and precision of λ in the presence of complex structures.
Keywords
Cone-beam CT, Image Analysis, Registration
Taxonomy
IM- Cone Beam CT: Registration
Contact Email