| |||||||||||||||
|
Home
|
For other articles and previous issues click here. January 30, 2006 1-2
Punch — NIH Study Data Supports CT Colonography With CAD’s
Value Clinical study results matter to scientists. The data from a large-scale National Institutes of Health (NIH) study is clarifying the performance of computed tomographic colonography (CTC)—also known as virtual colonoscopy—combined with computer-aided detection (CAD) technology. Virtual colonoscopy supporters hope the growing data will increase acceptance of the minimally invasive virtual colonoscopy in the eyes of patients and physicians. Results from the NIH study were presented at RSNA 2005 by senior investigator Ronald M. Summers, MD, PhD, chief of the clinical image processing service and chief of the virtual endoscopy and computer-aided diagnosis laboratory at the NIH Clinical Center in Bethesda, Md. “Our research took CTC CAD over a threshold,” Summers says. “The results suggest that it does indeed perform very well at finding polyps. Also, we were able to clarify the polyp size ranges for when it works most efficaciously. Specifically, the NIH study demonstrated that CAD used as a second read in conjunction with CTC showed almost the same sensitivity as optical colonoscopy for detecting polyps. A
New Threshold “We showed that CTC CAD really is feasible for screening patients and that it can perform quite well,” he says. CTC supporters hope the test’s evolution would not only increase radiologists’ diagnostic confidence in the exam, but that it would also increase screening compliance. The American Cancer Society recommends that people should begin regular screening for colon cancer at the age of 50, but compliance rates have been distressingly low. Much pain and suffering—and death—could easily be avoided if the screening population adhered to regular examinations. Colorectal cancer, the third most common form of cancer diagnosed in the United States, takes a long time to develop, providing a large window of opportunity to detect it early. It can take as long as 10 years for adenomatous polyps to develop into cancer. Such a long incubation period offers an excellent opportunity for early detection. Studies show that colon cancer detected early can be successfully treated in most patients. However, only approximately one third of colorectal cancers are found at early stages, simply because people choose to ignore screening guidelines. Many people simply don’t want to undergo colonoscopy, which they consider an ordeal. Virtual colonoscopy
could help boost compliance in the screening population because
it is a more patient-friendly procedure. CTC itself is minimally
invasive and poses no risk of bleeding or colon perforation. (Traditional
colonoscopy is performed to remove any polyps found, and the patient
is referred to oncology if cancerous lesions are detected.) Validation of CTC’s efficacy has come from smaller studies that have shown it to be a safe and accurate alternative to colonoscopy for detection of polyps. Also, subjects in these studies indicated that they prefer CTC over conventional colonoscopy. CTC
Challenges Summers believes CAD improves the performance of CTC by reducing perceptual error. With CAD technology, after the radiologist has interpreted the images, the computer acts as a second set of eyes, reviewing the images and marking abnormalities for the radiologist to then review. “The idea is that the radiologist would read the scan and then refer to the CAD detection and then re-review the sites that CAD pointed out and finally arrive at a diagnosis,” Summers explains. “Typically, I expect that a radiologist would not change his or her initial diagnosis, but if CAD pointed out something that the radiologist had overlooked, then the radiologist might add that detection to their diagnosis.” Mammography commonly uses CAD for a second read. Applied to CTC, CAD characterizes polyps according to distinctive features and by distinguishing detected sites such as polyps or false positives. One way CAD accomplishes this is by surface shaping, which identifies and characterizes a polyp using a curvature assessment algorithm to define its shape. Typically, colonic polyps have a rounded contour and an elliptical peak shape. CAD also involves CT attenuation to help distinguish polyps from false positives. False positives typically have low CT attenuation and polyps tend to have soft tissue attenuation. The
NIH Study “A number of small studies have shown that CTC with CAD is effective at finding polyps, so my goal was to do a large study to show a true expected performance of CAD, to get a better sense of whether this was near to clinical utility,” says Summers. The patients were screened at Walter Reed Army Medical Center in Washington, D.C., the NIH campus in Bethesda, and the Naval Medical Center San Diego. All underwent CTC and optical colonoscopy on the same day. The computer-aided CTC found 89.3% of polyps greater than 1 centimeter in size and 85.4% of polyps 8 millimeters or larger. Optical colonoscopy found the same percentage of polyps greater than 1 centimeter. However, CTC with CAD didn’t perform as well as standard colonoscopy for smaller polyps. With polyps that were 8 millimeters, the sensitivity of CTC with CAD measured only 76.4%. Standard colonoscopy demonstrated a significantly higher sensitivity of 90.9%. Still, Summers was pleased that CTC with CAD worked so well with polyps that were 10 millimeters or larger. He described those polyps as the most worrisome because of their size. False-positive rates for CTC with CAD were 2.1 per patient for polyps 10 millimeters and larger and 6.7 per patient for polyps 8 millimeters and larger. Those figures, the researchers point out, fell within an acceptable limit. The researchers concluded that the sensitivity and false-positive rates of computer-aided adenomatous polyp detection in an asymptomatic screening population were in the range likely to be clinically acceptable at both 8- and 10-millimeter size thresholds and were generalizable to fresh virtual colonoscopy data. Summers believes it’s likely that CAD will become “mainstream technology” used by all physicians who use CTC. “I was pleasantly surprised at how well we could get the CAD to work, particularly that it performed comparably to optical colonoscopy for polyps a centimeter or larger,” he says. Comparing
CAD Systems For the study, researchers constructed a library of 65 clinical CTC datasets with complete colonoscopy correlation (31 true positive and 34 false positive by prospective radiologist assessment). An unblinded radiologist not participating in the CAD analyses documented the exact location of every endoscopically verified polyp. The datasets were evaluated by two scientific teams using both CAD systems. Team members were blinded to all endoscopic results. CAD detections were compared with the previously specified reference standard. Results revealed
that Siemens’s system had a sensitivity of 73% (16/22) for
detecting lesions greater than or equal to 1 centimeter, while the
Summers’s CAD tool had a sensitivity of 95% (21/22) for lesions
similar in size. According to the study, the Siemens CAD had 0.6
false positive detections per patient, while the NIH CAD had 4.6
false positive detections per patient. Of the six polyps greater
than or equal to 1 centimeter missed by Siemens CAD, four were flat
polyps, a morphology for which their system is not optimized. Break
on Through But clinical application was inhibited by certain inherent problems, including the potential for errors on the part of readers, long reading times required—that CAD appears to be effectively addressing. Observers believe that as CAD systems evolve, the technology will carry CTC to greater acceptance in a clinical setting, for both patients and physicians. Currently, Summers can’t predict how long it will take for CAD to bring CTC into widespread usage. But after hearing him discuss the results of the NIH study, a listener can’t help but get the sense that Summers believes it will happen. —
Dan Harvey is a freelance writer based in Wilmington, Del., and
a frequent contributor to Radiology Today.
|
 |
|
3801 Schuylkill Rd • Spring City, PA 19475 Publishers of Radiology Today All rights reserved. |