Flowchart of the proposed method.
Coronary tree mask used in image coregistration. (a) Presegmented 3D binary mask covers the coronary tree region. 3D mask is shown with a representative 2D slice. This representative slice from 3D image is overlaid with presegmented mask boundary (line) (b). The 2D presegmented binary mask for this slice is shown in (c), and the 2D soft mask for this slice generated by convolving with the Gaussian kernel is shown in (d). The 3D soft mask is created for the entire 3D coronary region.
Anatomical landmarks used for validation shown in one example case with images from a baseline scan at the top row and images from a follow-up scan at the bottom row. We have obtained independent sets of anatomical landmarks from two clinical observers. Crosses indicate the positions of landmarks, including the left main coronary artery origin (a), bifurcation of left anterior descending and left circumflex arteries (b), right coronary artery origin (c), and branch point of the first diagonal artery from the left anterior descending artery (d).
Plot of the total energy and the energy under the volume-preserving constraint. For the latter, we preserve the volume of the coronary tree explicitly in such a way that the volume of soft mask remains constant, while the flow is minimized. The energy of volume-preserving constraint is almost the same, while the total energy is minimized.
Example of global displacement and local deformation techniques. Global displacement model (a) resulted in 2.0 mm for TRE and subsequently local deformation (b) improved this result to 1.4 mm for TRE. The target image is shown in (c). The lesion is marked with arrows. The global displacement model performs similarly to the local deformation model in terms of quantitative measurements, but visual assessment in specific cases showed there is still error especially in small plaque areas.
Example of coregistration results. Baseline image is registered to the follow-up image. Transverse orientation is shown at the top row, and coronal orientation is shown at the bottom row. The original baseline image before (a) and after (b) coregistration is shown. Panel (c) shows the follow-up image. As shown in (a), the original image is significantly misaligned. The “roving window” technique in (d) using a portion of the registered image from (b) to the target image in (c) shows that the result from our method is visually accurate.
Clinical example of plaque lesion progression (worsening) in an 82-year old man. The top row shows a follow-up study and bottom row shows a registered baseline study. The time difference between the two scans was 11 months. The pixel sizes of baseline and follow-up study were mm and , respectively. Arrows show increased, noncalcified plaque and stenosis in follow-up study.
Clinical example of plaque without significant changes in a 59-year old man. Top row shows the follow-up study and the bottom row shows the registered baseline study. The lesion is marked with arrows. The time difference between the two scans was 12 months. The pixel sizes of baseline and follow-up study were mm and , respectively.
Patient characteristics and image parameters of baseline and follow-up scans.
TRE comparison with different coregistration methods .
Inverse consistency error.
Average landmark distances between two observers for each landmark .
Article metrics loading...
Full text loading...