The prototype interpretation software actually produced lower sensitivities than endoluminal views for medium- and large-sized polyps, while detecting more of the smallest polyps compared to fly-through endoluminal viewing. And significantly, the dissection software's sensitivity for all polyp sizes rose significantly among patients who underwent thin-section imaging on a 16-slice CT scanner, rather than a four-detector system.
Perhaps most encouraging, the prototype viewing system slashed interpretation time from 38 minutes down to about 10 minutes per case.
Different methods of reading virtual colonoscopy data are preferred by different readers, in fact, preferences and experience levels vary widely. Trial results have been mixed, and the results tend to reflect the readers' experience and preferences. Some readers prefer to start with 2D axial images or multiplanar reconstructions (MPR) of the colon data, using 3D endoluminal views to confirm detected polyps. Others prefer to start with endoluminal views and confirm the findings with plain axial images.
To find better viewing solutions, researchers are pouring their efforts into developing novel viewing systems, which attempt to improve sensitivity and specificity by making the colonic mucosa easier to read.
Endoluminal image rendering is favored by many readers because it provides a colonoscopy-like method of navigating the colon. However, radiologists cannot see behind the haustral folds, and as a result must navigate back and forth along the centerline to evaluate the colon adequately, lead researcher Dr. Markus Juchems said in a presentation at the 2005 European Congress of Radiology (ECR) in Berlin.
The retrospective study by Juchems and colleagues from the University Hospital of Ulm and Karl-Olga-Krankenhaus in Stuttgart, both in Germany, and the University of Maryland Medical Center in Baltimore tested a novel dissection or "filet view" VC interpretation software application that flattens out the colonic mucosa in a type of modified Mercator projection, thereby eliminating the problem of blind spots behind folds. The study has also been published in the online version of European Radiology (June 14, 2005) in advance of an upcoming print version.
"Filet view dissects the colon and unfolds it," Juchems explained in his presentation. "We have a certain overlap at the borders of your picture so that you have a 360-degree view in the middle of the picture, and you have a 10-degree overlap at the top and a 10-degree overlap at the bottom, so all in all you have a 380-degree view of the colon."
The filet view function was developed by Dr. Thorsten Fleiter, an assistant professor of diagnostic imaging at the University of Maryland School of Medicine in Baltimore; Jacob Durgan of Philips Medical Systems; and Doron Dekel of Claron Technologies. The viewing option is included in Philip's Virtual Colonoscopy Tool (v2.x) in the firm's Brilliance Workspace, which was released commercially in 2004. A beta version was used in the study.
Virtual dissection or 'filet view' of the colon (left), and virtual endoscopy image (right) are alternative methods of examining CT images of the colon. Images courtesy of Dr. Marcus Juchems.
"Vos et al (Radiology, September 2003, Vol. 228:3, pp. 878-885) obtained their 360° view by using an unfolded cube projection, which renders 6° x 90° images together," the study authors noted in European Radiology.
The pilot study sought to compare the sensitivity, specificity, and postprocessing time of the colon dissection approach versus 3D endoluminal viewing. The researchers examined CT data from a small cohort of 21 patients (9 men, 12 women; mean age 58.1 years, range 34-76).
All patients underwent a standard cathartic polyethylene glycol bowel preparation, followed by administration of 40 mg of hyoscine N-butylbromide (Buscopan, Boehringer Ingelheim, Ingelheim, Germany), and insufflation of room air using a manual balloon pump.
"The patient controlled the amount of filling and stopped air insufflation if discomfort was experienced," the team wrote. "Air filling and distension of the colon was evaluated on the CT scout before CT colonography. Additional air was insufflated in case of insufficient distension of the colon."
Prone and supine CT images were acquired on a four-detector scanner in 11 patients (Philips MX 800, Philips Medical Systems, Andover, MA). In the remaining 10 patients, CT data were acquired on a 16-slice scanner (Philips MX 8000 IDT) at 120 kVp and 150-200 mAs. The four-detector scans were acquired at 4 x 2.5 mm with a slice thickness of 3.2 mm; on the 16-slice scanner, collimation was 16 x 0.75 with a slice thickness of 1 mm.
A reader experienced in virtual colonoscopy (and the endoscopic but not the dissection software) interpreted the results from both display methods on a single workstation (Philips Extended Brilliance Workspace 1.2) in random order, blinded to the colonoscopy results. Parallel MPR views were available for further examination of suspected lesions on the endoscopic views only.
"The filet view did not reach the same high sensitivity like the endoluminal view, but for very small lesions scanned with high-resolution protocol, filet view was superior to endoluminal view," Juchems said. "And the filet view was superior in evaluation time of datasets. It took us an average of 10 minutes (range, 5-13 minutes) to evaluate the dataset with filet view versus 38 minutes (range, 23-55 minutes) with the endoluminal view."
In all, 37 lesions (35 polypoid lesions, one colon carcinoma, and one lipoma) were detected on conventional colonoscopy, which served as the reference standard. Overall per-lesion sensitivity using the colon-dissection approach was 47.1% for lesions smaller than 5 mm, 56.3% for lesions 5-10 mm, and 75% for lesions larger than 10 mm; compared to 35.3%, 81.5%, and 100%, respectively, for endoluminal fly-through viewing.
Comparing the results from the 16-slice scanner data only, endoluminal viewing revealed sensitivity of 33.3%, 80%, and 100% for lesions smaller than 5 mm, 5-10 mm, and larger than 10 mm, respectively. This compared with sensitivities of 66.7%, 60%, and 100%, respectively, for the dissection or filet viewing software. Overall specificities were 84.6%, 87.5%, and 100%, respectively, with endoluminal viewing; and 69.2%, 75%, and 94.1%, respectively, with the virtual dissection view.
Still, overall sensitivities were lower with virtual colon dissection rather than endoluminal viewing. The authors believe that the lack of MPR integration in the prototype software is the principal reason.
"In contrast to most commercially available endoluminal view software systems, it was not possible to further evaluate lesions detected with the dissection mode in respect to their location in the supine and prone lesions, for example," they wrote.
The significant interpretation time-savings was probably the most promising aspect of the trial, potentially enabling a high-volume workflow of virtual colonoscopy readings. One reason for the comparative speed of dissection is that the prototype software didn't provide a reliable automated path extraction requiring a manual navigation through the colon, the authors wrote. The endoscopic interpretation was performed in ante- and retrograde directions to enhance colonic surface visibility.
"The major contributor to the reduction in time for evaluation has to be attributed to the virtual dissection view approach," the team wrote. "This technique does not require ante- and retrograde viewing because a surface visibility of up to 98% is already obtained in a single direction mode."
The technique shows promise for the rapid evaluation of virtual colonoscopy datasets, and particularly very small lesions, the researchers concluded, adding that further studies with larger cohorts will be needed to confirm the utility of the viewing method.
"We are currently performing a large prospective study in our university which compares filet view to conventional colonoscopy findings," Juchems wrote in an e-mail to AuntMinnie.com. The current version of the software includes available MPR views and endoscopic views.
By Eric Barnes
AuntMinnie.com staff writer
October 6, 2005
VC path-planning technique maximizes observable mucosa, August 30, 2005
Colon CAD: VC's extra eyes face new challenges, August 5, 2005
Part II: Computer-aided detection identifying new targets, July 15, 2005
Cubism shapes the art of virtual colonoscopy, October 6, 2003
Copyright © 2005 AuntMinnie.com