Registration of central paths and colonic polyps between supine and prone scans in computed tomography colonography: Pilot study
An example of the comparison of (left-right), (anterior-posterior), and (superior-inferior) coordinates of supine (solid line) and prone (dashed line) colon centerlines as functions of distance from the anus.
An example of 3D view of prone (1) and supine (2) colon images, from one patient, segmented using fixed intensity threshold at .
An example of the distribution of LEPs along the axis for both prone and supine paths. LEPs are obtained by numerical differentiation following Gaussian smoothing.
Pseudo code that recursively matches LEPs for one axis. RL is the “recursion limit.” Initially, the function is called with . Since the arrays are passed between recursion levels by address, their contents can be modified and stored during recursion. At every recursion level, is updated and passed into the next recursion as an input parameter. The final value of is the output of this routine.
(1), (2), and (3) are frequency histograms of variable , variable , and variable , respectively, using 120 landmarks [for (1) and (2)] and 26 matched polyp pairs [for (3)]. Each subfigure gives the corresponding maximum likelihood fit (solid curves) for a half normal distribution. The two dashed lines above and below the fitted curve are the 95% bootstrap interval.
Illustration of the dynamic programming scheme for solving the optimal matching of S set (with ordered elements) and P set (with ordered elements). Suppose for cases (0)–(4) elements included in a polygon are optimally matched, i.e., the sum of the matching scores is maximized. When the strict ordering is preserved, the optimal matching for case (0) can only be one of the three cases (1)–(3), i.e., whichever case that has the largest total score. Case (4) is proposed when the strict ordering condition is slightly relaxed by allowing cross matching the two pairs of neighboring elements. Therefore, if the optimal solutions for the structures included in the polygons in cases (1)–(4) can be found, the optimal solution for case (0) is equal to one of the cases (1)–(4) that produces the largest sum of match scores.
Average (over 24 patients) mean misalignment distance vs number of iterations at five different recursion limits. Note provided the smallest misalignment over the widest number of iterations.
An example showing the coordinates vs path distance, to illustrate the pre- and postregistration correspondence between supine and prone paths.
Unregistered, base line-adjusted, and registered mean misalignment distances for all 24 cases.
Descriptions of the symbols used in representing the polyp registration results in Tables II–IV.
Results for matching only the polyps listed in the reference standard, using the polyp registration algorithm. The cases are numbered the same as in Fig. 9.
Results for applying the polyp registration algorithm to match the top 30 CAD hits from prone and supine data sets.
The average number of CAD hits per polyp in one of the two scans that would need to be viewed to match its correct counterpart in the other. Results are presented for sorting the CAD hits by polyp match scores and for sorting the hits by relative path distance.
Cases with missing colonic segments due to locally poor distension and/or cleansing. : length of missing colon segment(s); : total path length; MMD: mean misalignment distance following path registration. The cases are numbered the same as in Fig. 9.
Article metrics loading...
Full text loading...