Mammography has been gleaning the benefits of computer-aided detection (CAD) systems longer than almost any other modality. And, while the technology is still relatively new and definitely still developing, it has become a standard of care in the field.

A quick Google search will return a mound of literature touting the importance of increased detections. But as far as Tommy E. Cupples, MD, breast imaging and interventional specialist for ImageCare LLC (Columbia, SC), is concerned, that’s not where the focus should be. “With CAD diagnosis, we found both an advance in the tumor stage and size as well as an advance in the age of the cancer,” Cupples explains. “Both of those things, as opposed to detection rate, are a direct measure of something that predicts a better outcome for the patient.”

Taking place between January 1, 1998, and December 31, 2000, Cupples’ study started as a year-end audit to determine the effectiveness of ImageCare’s recently installed CAD system, the ImageChecker mammography system from R2 Technology Inc (Sunnyvale, Calif). Comparing radiologists’ performance on 7,872 pre-CAD screening mammograms to the same physicians’ results on 19,402 consecutive exams performed with the aid of CAD, the researchers quickly discovered a whole new way of measuring CAD performance.

Published in the American Journal of Roentgenology, the study showed that “detection rates for small, invasive breast cancers (1.0 cm or less) increased by 164%,” and these cancers were found in women “5.3 years younger than in those for whom CAD was not used.”1

“What’s better news for women is you prove you’re finding cancers earlier and at a more curable stage,” Cupples says. “If you find that small of a cancer in a woman who’s five years younger and that might not have been detected until another year or two down the road, then she also has a better outcome, a better chance for a cure.”

Staying Abreast on Innovation

With the growing implementation of breast MRI, the application of CAD is following suit. However, CAD for MRI serves a different purpose than when used in conjunction with mammography.

“With breast MRI, we’re talking about a whole different animal,” explains Craig Hanson, MD, medical director of the Seattle Breast Center and chairman of the Department of Diagnostic Imaging at Northwest Hospital & Medical Center (Seattle). “CAD used in MRI is related to contrast enhancement of breast tumors as a function of tumor angiogenesis. The enhancement is depicted as a color map overlay, which improves the radiologist’s ability to both detect and diagnose breast cancers.”

Hanson, who uses CADstream from Confirma (Kirkland, Wash) on every breast MRI case, found that CAD noticeably increased throughput.

“This product improved our efficiency considerably, because it does all of the postprocessing functions, and we can view images much more quickly,” he says, adding that on average, his practice generates 1,600 images for each breast MRI study. Once available, images are viewed on any workstation with Internet access.

What’s Good for the Goose …

Viatronix V3D Colon v.1.2, integrated with Medicsight ColonCAD, permits the user to interactively view colon CAD polyp identification marks during virtual endoluminal fly-throughs using real-time 3-D volume rendering.
Viatronix V3D Colon v.1.2, integrated with Medicsight ColonCAD, permits the user to interactively view colon CAD polyp identification marks during virtual endoluminal fly-throughs using real-time 3-D volume rendering.

In addition to mammography, the FDA has approved CAD for several other modalities, including one device for CT lung studies and one for dentistry use. Blazing a trail into other modalities, CAD has started making its mark with radiologists specializing in colon and chest studies.

For gastrointestinal (GI) radiologists, watching the evolution of CAD for colonography has been a lesson in patience.

“It’s been quite frustrating for us, because those systems just haven’t been available to buy off the shelf,” says Steve Halligan, MD, FRCP, FRCR, of the University College London, who?in collaboration with Medicsight USA Inc (Nashville, Tenn)?has been using the CAD algorithm commercially available on the Vitrea 2 workstation from Vital Images Inc (Minnetonka, Minn).

Many in the GI radiology community are eagerly awaiting CAD systems because there is an expectation that some of the difficulties inherent to interpreting CT colonography will be performed by the software. In studies currently under way at the University College London, Halligan’s team is finding that CAD for colonography provides benefits for both 2-D and 3-D studies.

“In comparisons against a primary 3-D read, sensitivity does not increase, because sensitivity is already very high,” he explains. Researchers are finding, however, that the reading time decreases, resulting in tangible workflow benefits. “What we have found is that although the sensitivity of a 2-D scan with CAD is similar to a primary 3-D read, reading time is faster with CAD.”

For many GI specialists, the biggest benefit is being able to finally get their hands on the CAD software. “There are really fascinating avenues of research around CAD that will become possible because more people will have them and we’ll be able to do studies so much more easily,” Halligan says. “For 10 years, this hasn’t been available. Now, it’s going to be like everybody’s birthday.”

Breathing a Little Easier

With more radiologists requesting continuous thin sections from chest studies, the false positives of early CAD systems for use with lung studies have been lowered. As with any CAD system, though, false positives still occur. However, for the most part, they are easily identified and don’t demand additional time from the radiologist.

“The programs have become very sophisticated; most of the mistakes are made on either minimal scarring in the lungs or features along fissures against blood vessels, where you just know those are just not real nodules,” explains David Naidich, MD, professor of radiology at the New York University Medical Center.

Becoming familiar with the system also helps minimize the increased interpretation time. “I find that once you understand the rhythm of reading with CAD, eliminating the false positives can be done fairly quickly as you peruse the image set,” says David S. Mendelson, MD, associate professor of radiology at Mount Sinai Medical Center (New York). Mendelson currently works exclusively with R2’s CT Lung v.2.0 CAD System, which boasts the ability to deliver computer-reviewed images directly to the facility’s PACS.

Today’s CAD programs designed for lung CT excel at more than simply finding really tiny nodules. “When looking closely at CAD data from most studies, it’s not just small nodules that are picked up, but larger nodules are identified by CAD, too,” says Naidich, who employs the 510(k)-approved Nodule Evaluation Tool available on the Leonardo Workstation from Siemens Medical Solutions (Malvern, Pa). “In particular, it finds larger nodules that are difficult to identify?those that are in and around the central pulmonary vessels.”

Mendelson and Naidich agree that another benefit of CAD is its capacity to provide 3-D volumes of nodules, making it possible for physicians to compare whether the nodules’ dimensions have changed since a previous study.

Marked for Future Success

Beyond existing applications, the future possibilities are wide open in terms of where CAD can contribute to the improved healthcare of patients.

“As long as you can acquire the data digitally in some way, then you’re dealing with data that the computer can understand,” Halligan says. “One of my developers is always saying, ?If I can see it, I can train a computer to see it,’ which puts us on the brink of all sorts of possibilities.”

Cupples concurs, envisioning a future where CAD assists radiologists in every specialty. “I think that as CAD algorithms become increasingly sophisticated, every digital modality will have some form of CAD running in the background,” he says.

The unstoppable growth of technology is sure to carry CAD along with it. In fact, Halligan believes there is a potential for a global interlacing of CAD systems.

“Like the Internet, you could connect computers all around the country, all around the world on a grid,” he says, noting that if CAD programs drew from hundreds of hard drives, there would be an exponential increase in computing power and accuracy of the algorithms. “If we look 10 to 15 years in the future, we might find that none of us need a workstation, because the software is held on the grid and the computations are completed faster than you can blink.”

When peering into the chest, Naidich envisions capabilities well beyond lung nodules to “looking at airways and airway pathology, even looking at bone pathology. CAD could potentially look at anything that’s going on in the chest.”

And Mendelson is eager for improvements to CAD systems working to identify pulmonary embolisms (PE). “Pulmonary angiography is a mainstay at most hospitals, with many medical centers doing a lot of PE studies,” he says. “It’s a very detailed study to look at, and clinicians would like the advantage of an effective second read.”

Without a doubt, CAD is here to stay. And by all accounts, that’s a good thing.

“In a perfect world, all studies would be reviewed by more than one professional, and the evolution of CAD might give us the ability to do just that,” says Lynn Gayden, MD, medical director of the Women’s Health Center at the Baptist Memorial Hospital for Women (Memphis, Tenn). “Having a study reviewed by a physician with training, experience, common sense, and knowledge of the pertinent clinical information, as well as a computer program that is built on unbiased evidence-based data, could offer superior diagnostic capabilities.”

References

  1. Cupples TE, Cunningham JE, Reynolds JC. Impact of computer-aided detection in a regional screening mammography program. Am J Roentgenol. 2005;185:944?950.

CAD COLLECTION

Looking for a CAD system to complement your facility’s needs? Here’s a roundup of products currently available.

Confirma (Kirkland, Wash)
www.confirma.com

  • CADstream?The FDA-approved CADstream enhances the efficiency and standardization of MRI studies by automating data analysis and providing a network-based workflow architecture, which assists radiologists in the analysis, reporting, and interventional planning of these data-intensive studies. CADstream’s automated features include image registration; multiplanar reformatting; subtractions; angiogenesis maps; curves; maximum intensity projections; volume summaries; portfolio for detailed reporting; and multimodality readiness, offering users comparisons for mammography, ultrasound, and breast MRI studies.

Eastman Kodak Co (Rochester, NY)
www.kodak.com/go/health

  • Mammography CAD?Kodak’s CAD system provides touch-screen interfaces and convenient bar-code readers to ensure a minimum impact on existing workflow. Digitizing film is easy?films can be scanned in any order, with up to five cases loaded at a time?and fast, taking just 2.5 minutes to create digital images from four standard-size mammograms. Results are forwarded to a reporting station (or paper printer) and include the patient label and a bar code. A variety of hanging protocols means that images can be displayed as desired.

iCAD Inc (Nashua, NH)
www.icadmed.com

  • ClickCAD?Ideal for breast imagers with only an occasional need for CAD services, this fee-per-procedure model makes it possible for every interested medical professional to access CAD?without capital investment or a long-term contract. A prepaid authorization key makes CAD available for as little as $8 per case. Once the key is purchased, iCAD’s Second Look 200 system (see below) is installed at participating locations without charge.
  • Second Look?iCAD’s series of Second Look systems all provide maximum sensitivity and optimal performance. Each system?Second Look 200, Second Look 300, Second Look 700, or Second Look Digital?is tailored to meet the needs of a facility’s volume and technical capabilities. Regardless of which product is selected, the Second Look makes it possible to detect actionable missed breast cancers earlier. Regions of interest are marked clearly and unobtrusively, and results can be customized to match established hanging protocols.
  • TotalLook?The TotalLook solution gives radiologists the big picture. By providing a way to conveniently and quickly compare film mammograms from prior exams to current study results, the system makes detection of subtle changes easier than ever. Currently available to digital mammography users on systems from GE Healthcare (Waukesha, Wis), Siemens Medical Solutions (Malvern, Pa), and Hologic Inc (Bedford, Mass), the solution is compatible with HL-7 and DICOM worklists and can digitize standard mammography films in 23 seconds (without any degradation of image quality).

Medicsight USA Inc (Nashville, Tenn)
www.medicsight.com

  • ColonCAD API?A recently announced partnership between Medicsight and Viatronix Inc (Stony Brook, NY) allows existing Viatronix V3D Medical Imaging workstations to be outfitted with Medicsight’s ColonCAD API image analysis software tool. Users can interactively view Medicsight’s ColonCAD polyp identification marks while viewing a virtual fly-through of the colon. Medicsight’s CAD tools are designed using a concurrent-read approach, meaning radiologists can view the original images simultaneously with Medicsight CAD findings.
  • LungCAD API?Also part of the partnership, users of Viatronix’s V3D Medical Imaging workstations will soon be able to add Medicsight’s LungCAD API image analysis software tool. Users can interactively view Medicsight’s LungCAD lung nodule identification marks while viewing 3-D CT lung images using real-time volume rendering. This system also employs the concurrent-read approach.

R2 Technology Inc (Sunnyvale, Calif)
www.r2tech.com

  • ImageChecker?The ImageChecker product line leverages a growing database of clinical cases to effectively identify malignant masses, architectural distortions, and microcalcifications in a diverse screening population. With a range of systems available, R2 can meet the needs of any office: analog or digital, high or low volume. Upgrading is easy, so investments do not have to come before capital. R2’s proprietary CAD solution also is available with all of the major full-field digital mammography systems as well as with the IDS5/mx.net Breast Imaging Workstation?a multimodality, vendor-neutral,universal mammography workstation from Sectra North America (Shelton, Conn).
  • ImageChecker CT Lung?A leader in CAD for lung CT, this system features R2’s proprietary CAD technology platform. Its Lung Nodule application examines CT studies in three dimensions, highlighting potential actionable lung nodules and quickly computing measurements of detected regions. Reports are populated automatically with both the radiologist’s and the CAD-detected findings. R2’s Lung CAD option is available on any DICOM-enabled PACS as well as the Vitrea workstation’s CT Lung visualization package from Vital Images Inc (Minnetonka, Minn).

Siemens Medical Solutions (Malvern, Pa)
www.usa.siemens.com/medical

  • CADVision?Acquired by Siemens Medical Solutions in July 2004, CADVision Medical Technologies (Jerusalem) brought with it a unique mammography CAD system. CADVision’s technology is currently undergoing clinical validation.
  • Nodule Enhanced Viewing (NEV)?Available as an enhancement to the syngo LungCare CT, Siemens Medical’s NEV software is intended to support the detection of pulmonary nodules in thoracic CT examinations. This computer-assisted second reader product automatically marks suspicious pulmonary lesions, which can then be analyzed with the advanced workflow tools provided in syngo LungCARE.
  • Polyp Enhanced Viewing (PEV)?Available on the syngo Colonography system solution, PEV is a computer-assisted second reader to support the detection of lesions in the colon. Combined with the other workflow-enhancing tools in syngo Colonography, such as automated polyp measurements, PEV is intended to help improve accuracy as well as reading efficiency.

Riverain Medical (Miamisburg, Ohio)
www.riverainmedical.com

  • RapidScreen?The first and only FDA-approved chest X-ray CAD system to help radiologists detect early-stage lung cancer, RapidScreen is designed to quickly identify solitary pulmonary nodules sized 9?30 mm and other suspicious nodules on the original standard AP/PA chest X-ray. RapidScreen offers film and filmless solutions as well as an upgrade path for those seeking to evolve from a film-based to a filmless workflow in the foreseeable future.

?DH


GOT QUESTIONS? The FDA Has Answers

Forget three little words. For most manufacturers, sweet nothings are delivered with just two: FDA approved. Hearing them doesn’t have to be an elusive dream. For the FDA (Rockville, Md), the benchmark is simple: Prove it works.

“CAD devices?at least the ones we’ve seen so far?are adjunct, intended to improve the sensitivity of the process,” explains Robert Phillips, branch chief of Radiological Devices for the FDA. “So all the manufacturer has to do is show that the person reading the image plus the device performs better than just the person alone.”

How is that done? Anyone in medicine already knows the answer: Conduct studies that test and demonstrate the device’s efficacy. The best place to start is to determine the indication for its use. Once a company clarifies exactly what role its creation will play, it can start developing an appropriate study to prove it is safe and successful.

Results coming out of these studies receive intense scrutiny from the FDA?and for good reason. As any statistician (and Mark Twain) will attest, not all numbers are created equal.

“You have to power your study to give you an answer with statistical validity,” Phillips says. Validity is based not on how many people participated in the research, but rather on the methods used to conduct it.

The absence of this vitally important data will quickly halt the review process. One example of such a stumbling block is data submitted for foreign-borne devices. The issue is not where the device is from; however, FDA representatives must see studies representative of the area in which the system is intended for use.

“That creates some problems when looking at foreign data, because the practice of medicine in a foreign country might not be the same as the practice of medicine in the United States, or the demographics of the population might not be the same,” Phillips explains. “It does not automatically disqualify foreign data; there’s no reason to disqualify them as long as the fundamentals are appropriate.”

Direct from the Source

Hockey great Wayne Gretzky once said, “You miss 100 percent of the shots you don’t take.” Action, it seems, has its own rewards. The same advice applies to manufacturers looking for approval from the FDA.

“The best thing a potential sponsor can do is talk to us,” Phillips says. “We’re happy to talk to companies at the beginning and at various stages of the process to try and point them in the right direction.”

Meeting with representatives from the Division of Reproductive, Abdominal, and Radiological Devices is the best way to find out if the application process?and all required supporting documents?is on the path to positive results.

As to be expected, there are established steps to getting on the FDA’s calendar. The first is to put together a meeting background package. Complete requirements can be obtained from the organization, but the package will include such things as objectives and outcomes expected from the meeting, proposed indications for the device in question, and specific questions to be discussed during the meeting.

“Getting the information before we meet means that time spent with the sponsor is more effective, because we can research various issues that the company has raised and can respond intelligently,” Phillips explains. If necessary, follow-up chats can be scheduled. The same premeeting protocol applies.

“It’s also important for sponsors to know that meetings don’t have to be face to face,” he adds. “They can be over the telephone or via video conference?we’re very flexible.”

It’s All on the Internet?Really

For CAD vendors?and CAD consumers, for that matter?who are interested in discovering just what the FDA likes to see, hop on to the World Wide Web and surf to the FDA’s “Device Advice” section (www.fda.gov/cdrh/devadvice). The site is full of such information as definitions of device classifications, the differences between a PMA and a 510(k), and guidelines for how products are classified.

“I also recommend looking in the databases,” Phillips says (www.fda.gov/cdrh/databases.html). “You can look at the labeling and the summaries of approved devices, which show the safety and effectiveness we sought.” The site also contains descriptions of the manufacturer’s studies, along with details on how the device performed.

Taking note of study techniques and labeling of products already on the market also can be enlightening. This information is housed on the Office of Science and Engineering Laboratories (OSEL) section of the site (www.fda.gov/cdrh/osel/). Also, several papers are posted on the recognition of consensus paradigm, a very elegant way sponsors can employ for demonstrating that the device works. They can be found on the Standards page (www.fda.gov/cdrh/stdsprog.html) of the OSEL section.

?DH


Dana Hinesly is a contributing writer for Medical Imaging.