A Comprehensive Cancer Center Designated by the National Cancer Institute

Pathways to Discovery: Winter 2011

CAD Improves Colon Cancer Screening

Each polyp detected with CAD that is initially missed by the radiologist is a potential life saved.
—Abraham Dachman, MD

Colorectal cancer is the second most common cause of cancer deaths in the U.S. An estimated 142,570 new cases and over 51,000 deaths are projected for 2010, yet only 50%-60% of the population 50 years or older have been properly screened as recommended by the American Cancer Society (ACS).

According to Abraham Dachman, MD, professor of radiology, “The state of Illinois is one of 21 states with a law that mandates private insurer coverage of colorectal cancer screening in accordance with ACS Guidelines. In 2008, the ACS Guidelines for colorectal cancer screening were updated to include computed tomographic (CT) colonography. Many local insurers are unaware of this law and as such, are not in compliance.”

CT colonography, also known as virtual colonoscopy (VC), is gaining acceptance as a method for screening the colon and rectum for polyps and masses. CT colonography is a noninvasive test that uses x-rays to produce multiple, cross-sectional images of the colon.

“Many radiologists have been slow to take advantage of training courses to learn how to read virtual colonoscopy exams,” said Dr. Dachman. A substantial learning curve exists, especially for novice radiologists, and the sensitivity of the method in detecting growth abnormalities can vary.

Dr. Dachman orchestrated a large clinical trial that examined the effectiveness of computer-aided detection (CAD) in improving the accuracy of interpreting CT colonographic images. The use of CAD systems, which are computer software programs designed to help radiologists identify abnormalities with greater accuracy and specificity, is increasing in both screening and diagnostic procedures.

“This was the most rigorously conducted CAD trial to date,” said Dr. Dachman. CT colonography exams were read by 19 radiologists of varying degrees of experience, both academic and non-academic, who searched for polyps on CT colonographic images from 100 patients known to have either negative or positive findings. The radiologists, who were unaware of whether patients had polyps or not, interpreted each image twice, once without CAD and once with CAD (using VeraLook software by iCAD, Inc.) in a randomized fashion. They noted the location, size, and structural features of any detected polyps.

The use of CAD improved the overall performance of radiologists. Thirteen of 19 radiologists showed an improved accuracy rate for detecting polyps. The average sensitivity for detecting polyps of 6 mm or larger improved by 11.8%. The detection of polyps that are 6 mm or larger is accepted as a key metric for evaluating the effectiveness of screening procedures for colorectal cancer.

Any improvement in the sensitivity of detecting polyps with CAD is clinically important. Patients who are found to have polyps during CT colonography usually undergo optical colonoscopy, the traditional method used for colon examination. Optical colonoscopy has an average miss rate of 6%, and based on CT colonography data, as high as 12% for 10-mm or larger polyps.

“In a screening program, even a single 10-mm or larger polyp that is missed by the radiologist interpreting the CT colonographic examination results is important because the patient may not undergo another screening CT colonographic examination for at least 5 years. Each polyp detected with CAD that is initially missed by the radiologist is a potential life saved,” according to Dr. Dachman and his colleagues.

“I hope that the addition of CAD software options will help improve reader confidence, encourage radiologists to learn how to perform and interpret these exams, and help lobby insurers to reimburse for screening VC,” said Dr. Dachman. “At UChicago, we have been offering VC to the public for over a decade, and our recent data show that about 60% of our exams have been reimbursed by insurance.”

These findings are published in the September issue of Radiology (Radiology 256:827-835, 2010).