Framework of the multiresolution strategy.
Locations of the landmarks (marked by spheres) at (a) TLC and (b) FRC for registration pair A. Surface rendering of five lobes and major vessel branches is shown for reference.
Distribution of landmark distance along the axis (from apex to base) (a) before and (b) after registrations using the SSTVD method (filled diamond) and the SSD method (unfilled circle) for registration pair A. Note that all landmarks have an approximately uniform error distribution from apex to base for the SSTVD method while they have a large deviation of error for the SSD method in (b).
Surface distance maps of the warped TLC major vessel tree to the FRC major vessel tree for the (a) SSTVD method and the (b) SSD method. The FRC major vessel tree is shown in white for reference. Note that the red segments in (a) are caused by missing segments in FRC major tree, not by mis-registration. Examples are marked as A and B. Segments with large errors in (b), marked as C, D, and E, are due to clear mis-registration, as one can easily see that the white unwarped segments misalign with the color coded tree.
A slice from the registration pair A. (a) Slice from the reference image. (b) Corresponding slice from the warped image with the SSTVD method. (c) Corresponding slice from the warped image with the SSD method. Notice a misalignment of vessel trees indicated by the arrow for the SSD method in (c).
A slice from the registration pair F. (a) Slice from the reference image. (b) Corresponding slice from the warped image with the SSTVD method. (c) Corresponding slice from the warped image with the SSD method. Notice the good alignment between the lobar fissures (highlighted in lines) and between the vessel trees for the SSTVD method in (b), while an obvious misalignment of the fissure and vessels for the SSD method in (c).
The TLV, AV, and TV for all six registration pairs are estimated from the HU based on quantitative CT measures. The percentage change in the tissue volume is also listed for reference. The unit for volumes is liter (l).
Statistical comparison of landmark errors between the SSTVD method, the SSD method, and the MI method for all six registration pairs.
Article metrics loading...
Full text loading...