CAD Closes In on the Colon
This clinical image of a colon was acquired using Siemens syngo colonography with PEV (polyp enhanced viewing) software.

The myriad twists and turns of the colon make it an area ripe for tools that help clinicians get better pictures and a helping hand in detecting suspicous areas that could be cancers and polyps, the precursors of colorectal cancers. Numerous studies around the globe indicate that non-invasive CT colonography (CTC), first introduced in 1994, is as effective as conventional colonoscopy. Colorectal cancer is the second leading cause of death from cancer in the United States and polyps in the colon precede most colorectal cancers.

Mehdi Cadi, MD, a radiologist at la Pitie Salpetriere Hospital in Paris, and his colleagues have been performing CT colonography since 1999. After completing more than 1,500 exams, they added LMS-Colon CAD software from Median Technologies to improve their performance. A CAD algorithm flags suspicious regions, which guides the radiologist to review a spot that may have been overlooked during the initial analysis—hopefully increasing early detection.

Cadi and his colleagues evaluated the impact of CAD on radiologist detection of polyps in CTC. They evaluated performance by a junior radiologist without experience and a senior radiologist with CTC experience. The images were analyzed with and without CAD. The overall sensitivity of virtual colonoscopy without CAD was 52.6 percent compared with 73.3 percent with CAD for the junior radiologist, and 89.4 percent compared with 94.7 percent for the senior radiologist. “We concluded that CAD increases sensitivity in detection of colonic polyps and the effect of the CAD was greater with the beginner reader and improved the learning curve of junior radiologists.”

Cadi has organized several CTC workshops and spoken at several major clinical meetings and says the most common question concerning colon CAD he is asked is whether it should be used as a first or second reader. “We strongly recommend using this tool as a second reader,” he says.

Testing, testing

Thomas Mang, MD, a radiologist at the Medical University of Vienna in Austria, agrees that CAD should only be used as a second reader. He has used Siemens Medical Solutions syngo Colonography since 2001. “We had access to the CAD prototype in 2005 and spent a lot of time testing the algorithm,” he says. The facility collaborates with Siemens on research and practical work. “CAD potentially improves the reading performance of the radiologist and reduces perception errors in data evaluation.”

However, the “CAD algorithm cannot compensate for training and expertise,” he says. “The radiologist has to decide if the finding is true or not.” He says radiologists should perform a full, unassisted evaluation and then assess the CAD findings. The effect of CAD algorithms depends on the skill of the reader. An experienced reader will find most of the lesions himself or herself, as Cadi’s research indicated, while a less experienced reader may profit from CAD more. However, the correct interpretation of a CAD finding has to be done by the radiologist.

Most CAD algorithms are assigned to detect colon polyps ranging from about 5 to 25 millimeters. Larger polyps are easily detected without CAD and smaller lesions have a low risk of developing into cancer.

Mang organizes a CTC workshop in collaboration with Siemens in Vienna several times a year to train radiologists and residents in using the workstation and reading the cases. All participants have to perform guided evaluation of about 20 cases to get basic experience in using the workstation and perception and interpretation of colonic lesions.

Coming soon?

In the future, Mang thinks that increasing the sensitivity and reducing false-positive findings of the CAD algorithm are important, “especially common false positives like the ileo-coecal wall and the rectal tip.” The requirement of a laxative for bowel cleansing is a primary barrier for patient compliance, and retained stool can simulate polyps, decreasing specificity. So, an algorithm that works with reduced bowel prep and fecal tagging is important, he says. Mang also looks forward to automatic matching of CAD findings in prone and supine positions.

Focusing on CTC software, “I am very interested in new, advanced imaging [techniques] like a panoramic view or colon flattening that makes 3D evaluation more time-efficient.”