CAD Makes its Mark on Diagnosis

Computer-aided detection (CAD) shows great promise for assisting radiologists as they analyze images to detect cancers of the breast, lung and colon. A CAD algorithm highlights and marks suspicious regions, which steers the radiologist to review a spot that may be overlooked during the initial analysis. In theory, CAD boosts cancer detection rates. This month, Health Imaging & IT asks how CAD pans out in practice.

Breast CAD, which first earned Food and Drug Administration approval in the summer of 1998, is the most mature technology and available for both analog and digital mammography. Lung CAD comes in two flavors-x-ray and CT-based solutions. CAD systems for lung x-rays circle suspicious nodules on the classic chest film, while CAD systems examining CT images aid radiologists by marking suspicious areas and completing automatic measurements as they navigate through the tremendous volume of data generated by a chest CT. The newest CAD option - colon CAD - is based on a similar premise. That is, CAD helps the radiologist sort through the volume of data acquired during CT colonography and by placing marks on suspicious areas.

Breast CAD: Tried & true

Breast CAD is a winner. With fewer breast imaging centers and radiologists to practice mammography over the last several years, as well as high malpractice rates, breast CAD lends mammographers a helping hand with great skill. "CAD certainly helps radiologists diagnose more cancers. All studies across the board from private practices to academic medical centers to experienced and novice mammographers demonstrate that CAD helps find cancers earlier and better," sums Rachel Brem, MD, director of breast imaging and interventional center at George Washington University Medical center (Washington, D.C.). Brem and her colleagues rely on iCAD's SecondLook CAD solution to analyze both analog and digital mammography images.

Multiple clinical studies link the use of CAD with an increase in the breast cancer detection rate. Increases range from just over 6 percent to nearly 20 percent depending on the study and setting - with the average increase in the 15 percent range. For example, a two-year prospective study published in the American Journal of Roentgenology in October 2005 showed that CAD increased the cancer detection rate by 16.1 percent. This was the expected result, says chief investigator Tommy E. Cupples, MD, breast imaging and interventional specialist with ImageCare, LLC in Columbia, S.C.

Cupples says the study also revealed some unexpected findings. Studies that only look at the cancer detection rate don't tell about the nature of the cancer. That is, is it the size of a pea or a golf ball? Is the CAD finding smaller cancers earlier in the disease process?

"CAD is supposed to find small, irregular masses that are likely to be overlooked and associated with earlier cancers. So the real question is does CAD find small, earlier cancers," explains Cupples.

Cupples, who uses R2 Technology, Inc.'s ImageChecker, found that CAD does help find smaller cancers earlier in their development. The study found a 164 percent increase in small, invasive cancers less than one centimeter in size. In raw numbers, the results translate into two or three more cancers in a screening population in one year's time, estimates Cupples.

The study also examined patients' age at the time of diagnosis and found that the average patient age at time of cancer diagnosis decreased by 5.3 years with CAD. "CAD is doing exactly what is predicted-advancing diagnosis by finding subtle, faster growing cancers," states Cupples. A five millimeter mass in a 40-year-old woman is a larger concern than a similar mass in an 80 year-old patient. That's because the mass may increase only one or two millimeters per year in the older patient. The younger woman's mass, however, could swell to three centimeters in just one year.
Cupples' work relied on R2's ImageChecker in conjunction with analog mammography. Although there is no diagnostic difference with the digital mammography/CAD combination the digital option could be powerful in other ways. CAD can be more easily integrated into the digital mammography screening process primarily via the elimination of the extra step of digitizing each mammogram.

Cupples anticipates a bright future for breast CAD. For example, the digital mammography/CAD combination can facilitate quality assurance by reducing the impact of under and over-exposed films and motion blur.

Brem points to a recent New England Journal of Medicine study showing a 45 percent reduction in breast cancer mortality from 1975 to 2000 based on screening and treatment programs. "CAD gives us the ability to do even better," claims Brem.

Other studies are less optimistic. For example, researchers at St. George's Hospital and the Royal Free and University College Medical School, in London, found that CAD increases sensitivity of single reading by 1.3 percent from 90.2 to 91.5 percent. On the other hand, double reading increases sensitivity by 8.2 percent to 98.4 percent. Double reading, however, may not be economically or practically feasible given the shortage of radiologists and current mammography reimbursement rates.

Lung CAD: Now available for prime time?

Lung CAD solutions aim to solve an essential dilemma. "If you are in the business of looking for small lung nodules, you occasionally will miss one," explains David Mendelson, MD, associate professor of radiology at Mount Sinai Medical Center in New York City.

The other tricky part of lung cancer detection and treatment is its dismal survival rate. The American Cancer Society says the five-year survival rates in the United States for all stages of lung cancer is 15 percent. If lung cancer is found and treated while it is localized, survival rates increase to 49 percent. Currently, only 16 percent of lung cancers are found in the early, most treatable stages.

X-ray-based lung CAD systems, like Riverain Medical's RapidScreen, are very similar to mammography CAD solutions. These systems digitize the ubiquitous chest x-ray and apply a CAD algorithm to note suspicious nodules. Garden City Hospital in Garden City, Mich., has used RapidScreen for more than a year and currently examines chest films of high-risk patients - factory workers, smokers, secondhand smokers and asbestos exposed patients - with CAD. "It's found a few small nodules that we've missed. Just those few are significant," sums Jim Williamson, administrative director of radiology.

Other lung CAD proponents advocate universal application of the algorithm. "That's because those who are least likely to have lung cancer are most likely to have small nodules," says Williamson. For example, it's not unheard of to find a small malignancy on the pre-op chest x-ray of a low-risk patient.

The next leap with x-ray CAD is the digital x-ray CAD combination. "The transition to digital [x-ray] will be an incredible improvement. We can improve images before sending them to CAD, and everything [including CAD marks] will be reproducible," says Williamson.

University of Iowa Hospitals and Clinics in Iowa City, Iowa, has used EDDA Technology's IQQA Chest CAD to help detect cancer on digital x-rays for more than a year. The system is designed to look for nodules in the 5 to 15 millimeter range. "We use it as a catch mechanism on about 10 studies a day," says Edwin van Beek, professor of radiology. "It's very difficult to examine images with the right level of suspicion all the time, and it's easy to overlook something small."

Van Beek relies on IQQA Chest when viewing chest films where lung nodules might be present, such as the initial exam of a smokers or routine studies of cancer patients. "Sometimes it picks up something I saw but wasn't sure about, making it more likely that I will order additional tests. It doesn't need to pick up many additional cases because it does not require a major investment of time or effort."

IQQA Chest aids radiologists with semi-quantitative features. After the radiologist draws digital lines around a nodule, the system gives its dimension and location and generates an automated report. The ability to combine a CAD algorithm with tools to quantify the response to therapy is the direction of the future, says van Beek.

The other chest CAD option is the CT-based solution like R2's ImageChecker CT Lung. Mendelson says, "My experience matches the literature. R2 occasionally finds a nodule that I've missed." In fact, the occasional miss may be a somewhat routine occurrence. A recent study published in Chest indicated that 33 percent of patients whose CT exams were initially interpreted by radiologists as normal, without suspicious focal lung lesions at routine clinical reading, had significant lung lesions detected by ImageChecker CT CAD.

The additional CAD pick ups do not change the course of treatment for patients, but do provide more accurate assessment of the nodule, says Mendelson. A more accurate initial assessment translates into increased accuracy downstream over the course treatment.

Enhanced accuracy is made possible by automatic measurement and comparison tools to track nodule changes over time. "This feature does help decide which options are most appropriate for a certain patient population," says Mendelson. For example, some patients may be better served by palliative care rather than surgery.

Colon CAD: An up and coming option

Virtual colonography or CT colonoscopy is not new to Abraham Dachman, MD, professor of radiology at the University of Chicago. Dachman has been researching virtual colonography for more than a decade and uses multiple software packages including solutions marketed by GE Healthcare, Philips Medical Systems and Vital Images. In addition, researchers at the university patented a software program, which has been licensed to several companies. "A wide spectrum of studies for colon CAD shows that its sensitivity overlaps that of good human readers," says Dachman.

In fact, CAD can outperform even experienced readers. A recent study published in the American Journal of Roentgenology showed that CAD identified 81 percent of polyps compared to an average sensitivity of 70 percent for three expert readers. CAD detected 92 percent of polyps larger than nine millimeters. All polyps missed by experts were detected by CAD. The study's author, Steve Halligan, MD, professor of gastrointestinal radiology at University College Hospitals in London, used Medicsight's ColonCAD with Vital Images' Vitrea workstation and a stand-alone workstation.

The next step is to determine how CAD affects the observer's output, says Halligan. He designed a study that had 10 radiologists read 120 virtual colonography datasets without CAD and then read the same datasets two months later with CAD. The aim was to show how CAD changes diagnostic performance. The study showed that sensitivity increased 39 percent with CAD.

Italian researchers also report enhanced sensitivity with colon CAD. Daniele Regge, MD, head of the Institute for the Research and Treatment of Cancer (IRCC) of Candiolo, Italy, says preliminary findings with iMED's CAD Colon demonstrate that the algorithm is 100 percent sensitive for significant polyps greater than 1 centimeter compared to a human sensitivity of 85 percent for similarly sized polyps. The algorithm is 93 percent sensitive for six to nine millimeter polyps, while radiologists' sensitivity hovers in the 60 percent range.

Finally, Halligan has studied primary 3D reading without CAD to 2D reading with CAD and found comparable sensitivity, but the 2D/CAD model is faster. Dachman argues that readers need both 2D and 3D on every case with one interface used for reading and the second for problem-solving. CAD can be applied to any interface.

Speed, however, may not be the most critical challenge surrounding virtual colonography. Learning to read virtual colonography studies is difficult, and there is a long learning curve, says Dachman. "CAD can help make virtual colonography more sensitive and easier to read among both expert and novice readers. It can have its greatest impact with novice readers."

"Multiple virtual colonography CAD products will hit the marketplace in the next six months to two years. It's only a matter of time before there is an excellent colon CAD solution," predicts Dachman. "A good program will have a major impact. It must be sensitive with a low false positive rate."

The future of CAD

As new colon CAD systems earn FDA approval, radiologists around the globe are learning more about its benefits and applications. Halligan, for example, is studying how colon CAD works in a 3D reading paradigm.

Farther in the horizon, colon CAD solutions may perform like lung CAD systems. That is, they could incorporate tools to facilitate patient follow-up such as more accurate measurement tools and lesion characterization models.

Colon CAD is not the only option in fast and furious research mode. Cupples predicts, "As radiology goes digital, we'll use CAD for almost everything."

One of the next advances may be a pulmonary embolism CAD algorithm. Current options are plagued by a fairly high rate of false positives, but researchers are refining the algorithm. "If the algorithm is refined, it will be bigger than lung CAD," opines Mendelson.

New CAD algorithms aren't the only research focus. University of Iowa researchers aim to give clinicians a helping hand with a software suite to develop and quantify lung diseases like emphysema and pulmonary fibrosis.

Vendors continue to refine the granddaddy of CAD - breast CAD - with tools to indicate microcalcifications, varying sized marks which correlate to lesion significance and flexible operating points between high sensitivity and low false positive rates for microcalcifications and masses. And physicians are acclimating to these visual cues.

CAD will continue to demonstrate its utility, improving detection rates of cancers of the breast, lung and colon. CAD will spread with other applications such as the detection of pulmonary emboli and lover masses. And as CAD becomes further entwined with advanced visualization solutions, it will aid clinicians throughout the enterprise during the course of treatment with tools to quantify and track masses. CAD will modify and even swap out the D in detection for the D in diagnosis at some point, too. But there's a lot of diagnostic confidence to be gained between then and now. For sure, CAD is the wave of the future.