| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Full paper |
1 Department of Radiology, Academic Medical Center, Amsterdam, 2 Philips Medical Systems, Best, The Netherlands
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and methodsResultsDiscussionReferences |
|---|
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionReferences |
|---|
Since reference-points (e.g. hepatic flexure) are often not fixed, manual verification of findings on supine and prone positions may be a time-consuming activity. Using an automated supine-prone matching algorithm may facilitate this process, and may lead to a more efficient interpretation of CT-colonography.
The method evaluated in this article is based on similarities in direction of the centre-line and allows for compression and extraction of the centre-lines in both positions.
The first aim of the study was to assess the feasibility of automated matching of supine and prone CT-colonography examinations with colonoscopically proven polyps as fixed points of reference. Therefore the match error was calculated. The visibility of matched polyps was determined as well. This was carried out in a three-dimensional setting after a match with this new method was calculated. Third, the method was compared with a method that relies on the relative location between start and end of both paths.
|
Materials and methods |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionReferences |
|---|
5 mm) were used. These examinations formed part of a comparative study of CT colonography and colonoscopy in a surveillance population [1]. All polyps could be identified three dimensionally in both prone and supine positions. The polyps in both positions were linked, based on the three criteria of size, location and morphology. Polyps were excluded in cases where there could be doubt about the accuracy of these links. After manual insufflation of colon and rectum and intravenous administration of either 20 mg butylscopolamine bromide (Buscopan; Boehringer-Ingelheim, Ingelheim, Germany) or 1 mg glucagon hydrochloride (Glucagen; Novo Nordisk A/S, Bagsvaerd, Denmark), patients were scanned with a four-slice CT scanner (Mx8000; Philips Medical Systems, Best, The Netherlands) in supine and prone positions. Scan-parameters were as follows: 120 kV, collimation 4 mm x 2.5 mm, rotation-time 0.75 s, pitch 1.25, slice-thickness 3.2 mm, reconstruction interval 1.6 mm and reconstruction filter C. The tube-current varied between 25 mAs and 70 mAs depending on the waist circumference of the patient.
Following scanning, centre-lines were calculated semi-automatically; a "seed" was placed in the proximal colon and discontinuities of the colon were bridged manually. For calculation of the centre-line, software was used to reconstruct the colon three-dimensionally (Easy Vision; Philips Medical Systems, Best, The Netherlands).
The matching of prone and supine was performed based on two principles:
In order to reach maximal alignment, the outcome of the following function needs to be minimum:
|
|
Second, since corresponding directions of both centre-lines are often not situated in exactly the same part of the colon, expansion and compression of the paths was applied. To avoid matching of the hepatic flexure in supine position to a curvature in the sigmoid colon in prone position, a penalty-value for expansion and compression of the paths was applied. This penalty value was proportional to the expansion or compression used. Therefore, the more the centre-line was manipulated in order to match parts with a similar direction, the higher the penalty value was.
The sum of the outcome of the function of maximal alignment and the penalty value of the expansion/compression formed the match cost.
With the use of dynamic programming [2], the match cost was computed for each possible combination of points on the centre line in both positions. The combination of points was made such that the total cost (i.e. the sum of all individual costs) was minimal.
The performance of the algorithm was measured using the absolute match error (in millimetres). This match error was calculated by measuring the distance from the actual polyp location to the calculated (expected) polyp position, [A] (Figure 1
). This was done along the central axis in prone position after the match was made.
|
).
|
The match was considered successful if the matched polyp could be classified under category 1 or 2. This method was compared with a method based on the relative location between start and endpoint of this centre line (Normalized Distance Along the Centre-line, NDAC) [3]. Here, the beginning of the centre-line, the caecum, is located at index 0 on the path and the end, the rectum, is defined as index 1. All other points will have indices between 0 and 1. The NDAC absolute match error is then defined as:
Absolute match error = [NDACpolypsupine–NDACpolypprone] x Length Central Axis prone
The Wilcoxon-test was performed to test for differences in absolute match error in both methods. The null hypothesis was that there was no difference between both methods. With a p-value of less than 0.05, this hypothesis could be rejected.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionReferences |
|---|
5 mm that were visible in both supine and prone examinations. These 32 examinations included 53 polyps (5 mm or larger) visible in both positions. One polyp in two patients was excluded because of invisibility on the three dimensional display. These polyps in the rectum were both hidden behind the balloon of the inserted catheter and could only be seen in a two-dimensional read. One patient was excluded because 12 of the 14 polyps were situated in the rectum and sigmoid. Here there could be doubt about the correctness of the exact linking of the polyps seen in colonoscopy and colonography since many morphologically less specific polyps were seen. In total, 16 polyps in three patients were excluded.
Of the remaining 37 polyps, 26 (70%) were sessile, 8 (22%) were pedunculated and 3 (8%) were flat. 24 polyps were 5–9 mm, 11 polyps were 10–14 mm and 2 polyps were larger than 15 mm. These were both carcinomas (5%). Of the remaining polyps 14 (38%) were adenomas, 10 (27%) were non-adenomas and of the remaining polyps histology was not obtained.
Two polyps were situated in the descending colon and four polyps in the rectum. In the remaining four segments, 7 to 9 polyps were situated. Four of the 29 patients had undergone a hemicolectomy.
The median absolute match error was 14 mm (range 0–59 mm, average 20 mm) either proximal or distal from the actual polyp in prone position. In the observer study, 70% (26/37) of the polyps was directly visible in prone view. Of these directly visible polyps, 20 polyps (54%) were seen instantly without flying though the colon (category I), the remaining 6 polyps (16%) were also instantly visible, but the observer chose to move the virtual camera over a small distance to verify the polyp (category II). The remaining 11 polyps (30%) could not be seen instantly but all were found after a flight through the colon within 150 mm of the matching position (category III).
When the NDAC method was applied to our CT-colonography examinations, this resulted in a median absolute match error of 16 mm (range 0.5–105 mm, average 26 mm), slightly larger than the other method. There was no significant difference (p = 0.502) in match error between both methods.
Half way along the centre-line three polyps were visible, with a substantial difference in match error in favour of the method of maximal alignment of centre-lines (Figure 3). These polyps were all situated in the transverse colon. None of these three patients had undergone a hemicolectomy.
|
|
Discussion |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionReferences |
|---|
The overall difference in median absolute match error between both methods was small (2 mm), although half way along the centre-line there were polyps with substantial differences in match error (larger with NDAC). All these polyps were situated in the transverse colon.
The transverse colon is the largest intraperitoneally situated colonic segment [4] that can move relatively freely in the abdominal cavity. We assume that this is the most mobile part of the colon, based on our observations of the three-dimensional overviews in both positions. The fact that polyps with a substantial difference in match error are situated in this segment might suggest that the method that relies on maximal alignment of centre-lines has an advantage in this mobile region.
The methods of matching prone and supine positions were evaluated based on the assumption that the included polyps were (immobile) true polyps and, second, that the actual link between these polyps in both positions was correct. In case there was doubt about the status of the polyp and linking, the polyp was excluded. With these exclusion criteria a very robust reference standard was created. Consequently, using this reference standard we were able to determine most precisely the matching quality of the methods.
Matching is important in determining whether a suspected lesion is a polyp or faecal material. A rule of thumb is that a suspected lesion with an unchanged position in prone and supine is most likely a polyp, while a change of position is related to stool. An accurate matching algorithm is therefore helpful to reduce the number of false positive findings in an efficient way.
Pitfalls in diagnosis can be caused by adherent stool mimicking a polyp, in case faecal tagging is absent or inadequate. On the other hand, pedunculated polyps (large stalk) or polyps situated in mobile segments may move (or seem to move) when both positions are compared. Therefore, when assessing whether a suspicious finding is a polyp, it is important not only to take the location into account, but also homogeneity and morphology of the lesion.
Incorporated in a CAD algorithm, this matching tool could also be used to reduce the number of false positive findings. Näppi et al [5] used a region-growing scheme with distance calculations to divide the colonic lumen into overlapping segments that match in the supine and prone data sets. Polyp candidates detected by means of a CAD scheme were eliminated if they could be seen in only one of the two corresponding segments.
A limiting factor for generalizing the visibility of the matched polyps is the use of a 3D unfolded cube method for reviewing the data. Using this method, 70% of the polyps could be seen instantly when the match was made. This review method may have had a positive influence on the percentage of polyps that were seen instantly, since the unfolded cube display method creates a 360° view without major distortion of the image.
Although evaluated here in a 3D-setting, the method of matching prone and supine datasets can be applied in a 2D-setting as well. This can be done on condition that a centre-line is calculated through both corresponding colons.
Theoretically, the matching quality of the method of maximal alignment is not altered by absence of a part of the beginning or ending of a centre-line in one position (e.g. caecum or rectum). In other words, the absence of baseline adjustment, a condition in other clinically evaluated methods [3, 6, 7], will not have a large influence on the match error in this new method. This baseline adjustment may not be possible in patients with an inadequately distended or insufficiently cleansed caecum or rectum in one position. Although this condition was not present in any of the data sets used, this proposition was confirmed in a test case. Here we removed 50 cm of the beginning of the centre-line without large consequences for the match error in the remaining colon segments.
The design of this feasibility study, with selected polyps as fixed reference points, precludes an assessment of the clinical value of the matching tool. Further study should assess matching quality and time efficiency of the matching tool in a series of unselected CT-colonography examinations.
From this study we conclude that automated prone-supine matching of CT-colonography studies is feasible and the difference in median absolute match error from the NDAC method was small and not significant, although half way along the centre-line (transverse colon) more pronounced differences were seen in some cases.
For this research project, a grant from Philips Medical Systems was received.
Received for publication September 12, 2005. Revision received January 9, 2006. Accepted for publication January 18, 2006.
|
References |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  R. M. Summers, J. A. Swift, A. J. Dwyer, J. R. Choi, and P. J. Pickhardt Normalized Distance Along the Colon Centerline: A Method for Correlating Polyp Location on CT Colonography and Optical Colonoscopy Am. J. Roentgenol., November 1, 2009; 193(5): 1296 - 1304. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S PUNWANI, S HALLIGAN, D TOLAN, S A TAYLOR, and D HAWKES Quantitative assessment of colonic movement between prone and supine patient positions during CT colonography Br. J. Radiol., June 1, 2009; 82(978): 475 - 481. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
I. W. O. Serlie, A. H. de Vries, L. J. van Vliet, C. Y. Nio, R. Truyen, J. Stoker, and F. M. Vos Lesion Conspicuity and Efficiency of CT Colonography with Electronic Cleansing Based on a Three-Material Transition Model Am. J. Roentgenol., November 1, 2008; 191(5): 1493 - 1502. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| BJR | DMFR | IMAGING | ALL BIR JOURNALS |
