CAD moves out of mammography into everyday radiology practice

Vendors pursue quicker and easier workflow for multiple modalities

Computer-aided detection began modestly, confined to the single diagnostic niche of mammography screening. Now CAD is breaking out into disease areas other than cancer, covering additional anatomical sites such as the lungs, and moving into CT, MR, PET, and MR spectroscopy. In short, CAD is becoming an integral part of the day-to-day life of radiologists.

"Because it can be trained to find abnormalities that exist within the structures a radiologist is examining, we see CAD as more than just a cancer detection set of software tools," said Michael S. Klein, president and CEO of R2 Technology (Sunnyvale, CA). "We see CAD being used in the everyday diagnostic reading process of all kinds of examinations to detect not only cancer in its early stages but anomalies such as pulmonary embolism."

CAD was first commercially deployed for mammography screening in 1998 but, as with any new imaging tool, its use didn't take off until reimbursement materialized. This happened in January 2001, when Medicare reimbursement kicked in and other insurers followed suit. Today, approximately 390 insurance carriers cover breast cancer screening using R2's ImageChecker product.

Consequently, CAD for breast cancer screening is flourishing. R2 has installed about 700 ImageCheckers and has entered into development and marketing agreements with the major mammography vendors. In keeping with its two-year-old agreement to provide CAD software for GE's full-field digital mammography unit, R2 has shipped more than 150 ImageChecker products to GE Medical Systems to date, said Chris Joseph, a spokesperson for the company. R2 Technology recently entered agreements to supply its software to Siemens Medical Systems, Fischer Imaging, and Hologic/Lorad.

CADx Medical Systems (Beavercreek, OH) also has development and distribution agreements with GE, Fischer, and Lorad. The company, which received FDA clearance for its most recent Second Look software upgrade in May, has cut purchasing contracts with Consorta (Rolling Meadows, IL), a healthcare resource management purchasing organization for Catholic, faith-based, and not-for-profit health systems and CareCore National (Wappinger Falls, NY), a managed-care utilization management organization.

Both R2 and CADx continue to refine their products for breast cancer imaging and to increase availability. R2 is creating the Reference Library, which compares findings from a woman's mammogram against a database of hundreds of cases of biopsy-proved breast malignancies. CADx is developing a network that links small mammography centers with remote Second Look breast imaging sites for quick access to CAD analysis.

Newcomers to mammography CAD are releasing software programs that improve the processing of mammograms. In 2001, Scanis (Foster City, CA) unveiled the Mammex TR version 1.4 with an easy-to-use interface and lower false marking rate for the U.S. market. At the end of 2002, iCAD (Hudson, NH) introduced MammoReader II CAD, which incorporates two mammographic digitizers and a high-speed, dual-processor workstation that can handle as many as 39 breast cancer screening cases an hour. Distributing its products through Instrumentarium Imaging, iCAD also released software that increases by 25% the speed of digitizing and analyzing mammograms. The software automatically prints CAD findings.

Several companies are designing CAD systems for other imaging applications. Confirma (Kirkland, WA) is developing versions of its first product, CADstream, to assess tumor morphology and provide other diagnostic indices for breast MRI. R2 Technology is building on its OmniCAD technology to create an open architecture multimodality platform that will accept analog and digital images from any manufacturer and images from any modality. The company is adapting its core pattern recognition software to include a subtraction algorithm that will accommodate the reconstruction of multiple slices of CT lung imaging data. The work-in-progress ImageChecker CT software, which was shown at the RSNA meeting, identifies areas that may contain clinically significant lung nodules. It also measures and characterizes the lesions.

CADx announced a CT CAD product at the RSNA meeting. Scheduled for delivery in 2004, it will provide information about the size and shape of abnormalities in the lung. Deus Technologies (Rockville, MD), which has developed a digital CAD product for lung cancer, entered into a strategic partnership with GE for use of the software on GE's digital radiography systems. GE already has exclusive distribution rights to the Deus RapidScreen RS-2000, the only CAD software to receive premarket approval from the FDA for early lung cancer detection.

Down the road, R2's chief executive sees CAD helping in the analysis of functional data from PET and MR spectroscopy. The software will be tailored to show the correlation between spatial and anatomical relationships with biological processes such as metabolism and physiology, Klein said.

For CAD to become as natural as a viewing station in a reading room, companies will need to develop software that makes information management easier for the radiologist, he said. One workflow-related upgrade is an algorithm that adapts pattern recognition and the analysis of suspicious imaging findings to patient-specific information, such as a history of cancer or pulmonary embolism, exposure to environmental hazards, and demographics or lifestyle factors that increase the risk for disease. This should increase the likelihood of identifying potential regions of interest. Physician-directed CAD customizes disease detection algorithms to flag imaging findings that individual radiologists historically have had difficulty interpreting.

"Up until now, imaging technology has only been as good as the eyes and brain of the radiologist," Klein said. "CAD can help by zeroing in on the most suspicious areas and directing the radiologist's attention to areas that are most in need of further analysis."

To be successful, however, CAD has to improve the efficiency of the interpretive process, according to Klein.

"No one wants something that slows them down," he said. "CAD has to be fast and clinically relevant, so physicians will want to use it."