Computer-assisted breast imaging systems help find and characterize cancers

CAD software improves breast ultrasound, digital mammograms

New computerized systems that give doctors a “second pair of eyes” for looking at mammograms and other breast scans are showing great promise for detecting breast cancer, distinguishing it from benign problems without a biopsy, and tracking changes in a woman’s breast over time.

The systems, all developed at the University of Michigan Health System, are in various stages of readiness for clinical use. But UMHS researchers will report today that they have made significant progress in using computers to increase the accuracy of interpretation of digital mammograms and breast ultrasound images.

The team will present results from several of the computer-aided diagnosis, or CAD, systems at the annual meeting of the Radiological Society of North America (RSNA).

One of the most encouraging results shows that a CAD system improved the ability of highly experienced radiologists to tell cancerous tumors from benign growths on ultrasound breast scans. Such scans are often performed after a suspicious finding on a screening mammogram, to help determine if a biopsy is needed.

The new result was achieved using ultrasound images of 102 breast masses that were later confirmed as benign or malignant by a biopsy. No cysts were included. The three-dimensional images were acquired using a conventional ultrasound machine and a mechanical transducer-guiding system developed at UMHS. Then, they were analyzed by a CAD system, and read by five breast radiologists.

After the physicians scored each mass on a scale of the likelihood of malignancy, they were shown the score assigned by the CAD system to the same mass, based on algorithms for mass shape, shadowing, and border characteristics. The physicians altered their score about one third of the time, mostly in the direction that correctly reflected the nature of the mass on biopsy. In the clinic, these score changes might mean that fewer women with benign disease would be subjected to an invasive biopsy procedure.

Of course, no one thinks computers will take over from human doctors anytime soon. “In the near future, it won’t be possible for computers to replace radiologists for this kind of test, because a radiologist looks at the patient’s entire case, not just her ultrasound images,” says associate research professor Berkman Sahiner, Ph.D., who will present the results in a talk at RSNA. “But if radiologists work with computers, they could improve their accuracy and spare some women benign biopsies.”

Since the 3-D CAD ultrasound system helped improve the reading of highly experienced breast radiologists — who conduct ultrasounds on nearly every woman referred for imaging at UMHS for the evaluation of a suspicious breast mass — it may help less-experienced physicians even more.

The U-M’s CAD Research Laboratory team, in the Basic Radiological Sciences division of the Medical School’s Department of Radiology, has worked for more than a decade to develop CAD systems for breast imaging. Led by Heang-Ping Chan, Ph.D., they work closely with clinicians in the U-M Breast Imaging Division, led by Mark Helvie, M.D., to evaluate new techniques on images from U-M patients under research protocols approved by the U-M Institutional Review Board.

In all, the team aims to use computers to improve they way radiologists detect and interpret cancers on mammograms and ultrasound images. They also hope to spare women some of the “worried waiting” and additional imaging sessions that follow an abnormal mammogram, to make additional imaging for such patients as accurate as possible, and to minimize the number of women who endure a biopsy only to find out their condition is benign.

Several commercial CAD systems for reading traditional film mammograms to increase detection – but not classification — have been on the market for a few years. But Sahiner and his colleagues hope their CAD research will advance the field by improving mammogram classification of lesions as malignant or benign, enhancing breast ultrasound imaging, aiding with detection and classification on film and digital mammograms, and allowing precise tracking of an individual woman’s case over time.

The U-M team is also working on their ultimate goal: combining ultrasound and mammogram images to give a complete view of a breast mass. They’re developing ways to combine the two very different kinds of images digitally, and are collaborating with industry to develop a breast imaging machine that can acquire an ultrasound image and a mammogram simultaneously. The U-M team for this effort is led by Paul Carson, Ph.D., director of the Basic Radiological Sciences division.

U-M researchers will report findings from several of the projects at the RSNA meeting. Among them:

AD for detection of masses on digital mammograms:

As the popularity of film-free digital mammograms grows, the U-M team is working on a CAD system to analyze those images for better detection of suspicious areas. At the RSNA meeting, Jun Wei, Ph.D., will report that their four-part image analysis system was just as effective at detecting masses on 206 full-field digital mammograms as it was at detecting them on mammograms of the same breasts that were taken on film and then digitized. The U-M digitized-film mammogram CAD system has already been proven in previous studies and a pilot-scale clinical trial to give high detection rates, and the team has applied for a patent on the system. Now, the team plans further refinements to the CAD system to adjust it to the digital mammography images.

Tracking a woman’s breast health over time:

Almost as important as finding a woman’s breast cancer in the first place is monitoring her breast over time to catch further growth or change in suspicious areas. U-M researcher Ludomir Hadjiiski, Ph.D, will present results from 390 pairs of mammograms taken some time apart, and evaluated by five different CAD techniques for pinpointing the location and size of suspicious lesions. The highest level of accuracy was seen with a regional registration method developed by the U-M team.

Combining ultrasound and mammography in one scan:

Since women with suspicious screening mammogram findings are often referred for both additional mammography and an ultrasound, the U-M team is working with researchers from General Electric to develop a machine that can make both kinds of scan of the same area at once. At RSNA, they will report the results of initial design studies for a machine in which an automated ultrasound transducer is mounted above the compression paddle used to compress the breast for mammography. The results from studies on breast “phantoms” and on actual patients show that the paddle, made of a special plastic, doesn’t substantially affect the quality of the ultrasound image.