Facing the dual challenges of huge image data sets produced by computed tomography (CT) and the shortage of radiologists to read imaging studies, computer-assisted detection (CAD) holds the promise of riding the white horse to the rescue. While CAD for mammography has gained favor (and why not, enabling detection of 20 percent more breast cancers), a large push these days is directed toward lung nodule detection. And additional applications for CAD are not far behind.
Matthew Freedman, M.D., M.B.A., associate professor of radiology and clinical director of the imaging science and information systems research center at Georgetown University (Washington, D.C.) and medical director for Deus Technologies LLC (Rockville, Md), both an academician and someone engaged in the development of CAD applications in the private sector, suggests that these approaches are in the earliest stages of deployment.
My sense is that perhaps 10 years from now, the term CAD will be forgotten because it will be intrinsic in the tools that the radiologist uses, he says. If the radiologist is reading a chest x-ray, CAD will be inherent whether he or she is reading a soft copy or hard copy. It will become primitive not to use it. Freedman predicts that these tools will prove valuable across the spectrum of imaging modalities.
Steve Rogers, Ph.D., president, CEO and founder of CADx Systems (Beavercreek, Ohio) concurs. From a technological point of view, there is no reason that they cannot take any radiological data set and apply their algorithms to it. However, the key is finding those applications where CAD will add significant value either in the detection of abnormalities, or to increase the radiologists efficiency.
At the highest level, CAD is the same, independent of the applications, says Rogers. What CAD fundamentally consists of is a set of steps which include finding those sets of measurements which are unique to that modality to separate targets into different categories you care about, such as a solitary pulmonary nodule, and then apply the relevant context.
The CADx system shifts the paradigm from modality-dependence to a patient-centric approach through the use of their Qualia Insight integrated set of software tools that enable their engineers to develop CAD for any unique application. Originally, Rogers and Matt Kabrisky, Ph.D. (chief scientist emeritus at CADx), the founders of this company, helped the U.S. government to develop smart weapons. With each new demand, they would not be able to start development from scratch each time, so they designed an integrated set of tools to address all of the needs. Qualia Insight follows that systematic approach to designing software tools that can be applied to perform CAD functions regardless of imaging platform.
Lung Nodule Detection
Most of the activity in CAD development today is directed towards early detection of lung cancer in CT scans. Considered a works-in-progress by developers, all of whom are engaged in the FDA pre-market approval review process, CAD for lung nodule detection assumes a variety of approaches depending on the OEM involved.
Lung cancer is the No. 1 cancer killer in the United States (and worldwide) with projection of 157,000 deaths from this disease in the U.S. this year. As with other forms of cancer, early detection seems to hold the key to improved prognosis.
Diagnosis of solitary lung nodules presents a significant challenge to the most experienced and skilled radiologist in practice today. While imaging techniques offer the best non-invasive option for detection and diagnosis, the task can be equated to conducting the proverbial needle-in-a-haystack search.
R2 Technology (Sunnyvale, Calif.) will be releasing their CT product for the automatic detection of lung nodules in Europe in the third quarter of this year, according their vice president of CT products, Susan Wood, Ph.D. They have submitted the PMA to the FDA for ImageChecker CT LN-1000, which they showed as a works-in-progress at RSNA last year.
Within CT, we are concentrating on using our automatic detection algorithm as a springboard for a number of productivity enhancing steps to help the radiologist in an image intensive environment produced by multidetector CT, explains Wood. The sheer number of axial slices produced in these studies increases the difficulty in detecting solitary pulmonary nodules. The automatic detection component of this system provides a tool where the physician can sift through the rich imaging data more efficiently.
Peter Cooperberg, M.D., professor of radiology and vice chairman of the department of radiology, University of British Columbia (Vancouver, B.C., Canada) has been involved in development of the R2 product and explains the issues inherent in a CT lung study.
When he performs a screening chest CT, using their GEMS 4-slice LightSpeed scanner, it is difficult to discern between a blood vessel and a lung nodule when examining a single axial slice.
If you look at any image, there will be 100 little dots, some larger and some smaller. Virtually all of them are blood vessels, but a couple of them will be nodules, Cooperberg explains. If its a nodule, it may not be visible in one slice, appear in the middle slice and then disappear again in the slice above. Thats the difference between the blood vessel that would continue through all slices and the seed-like structure of a nodule.
Image of a segmented lung nodule in baseline and follow-up scan.
As they explore and refine the tool, he performs chest CTs in the usual manner, and then applies the CAD algorithm to look for additional nodules or confirm the ones he has found. The goal is to reduce the number of false positives or false negatives that the system identifies.
Wood reports that the clinicians who are currently assisting with development of the R2 system have provided feedback that the tool has improved their workflow. Given the huge quantity of data produced by a typical CT scan of the chest, accurate detection of specific nodules becomes an important activity.
Vital Images Inc. (Plymouth, Minn.) has entered into a partnership with R2 Technology for lung visualization CAD. This three-year agreement means that
R2s Image Checker CT lung nodule detection software will be integrated into Vitrea 2 software.
Jay D. Miller, president and CEO of Vital Images, explains that the large volume of high-resolution images produced, especially by the multislice CT scanners, can present real challenges to the physician trying to review those images manually. Coupling CT visualization with CAD provides the opportunity to find extremely small lesions.
Metin Gurcan, CT product manager for CADx, was charged with developing a CT lung nodule capability for their Qualia Insight technological platform. The prototypes for both lung nodule and colon polyp detection were demonstrated at last years RSNA.
Weve developed a platform so we can do CAD in any modality, whether its MRI, ultrasound or CT, explains Gurcan. Finding the right set of measurements, breaking things into categories and applying the context works whether its 1D, 2D or 3D. Although the algorithms are based primarily on three-dimensional data sets, it can be applied to any image data, and even to text-based information.
Gurcan explains that biopsy and resection of lung nodules is not only very risky but expensive as well. Therefore to avoid unnecessary surgical procedures (for benign lesions) the radiologist must be able to
measure the nodule, and examine its shape and characteristics, such as whether or not it is calcified and contains fat. Typically, the patient is imaged three months after the original scan to determine any changes that have occurred in the nodule.
In our platform, we have developed programs that will automatically find the same nodule for the radiologist, measure its characteristics such as size, and intensity and report those, says Gurcan. Because CT provides three-dimensional data, they are able to conduct true volumetric measurements.
Kabrisky adds that Gurcans measurements are extremely accurate and are able to determine such parameters as tumor doubling rate.
Alok Gupta, Ph.D., M.B.A., director of computer aided diagnosis and therapy for Siemens Medical Solutions (Malvern, Pa.), explains that their approach involves considering CAD from a multimodality perspective, and a disease perspective.
With both CT and MRI, radiologists perform soft-copy reads off workstation screens. With lung or abdominal imaging, hundreds of images may be generated and sometimes anomalies are subtle and not clearly visible. Gupta asserts that the primary beneficial role of CAD is to reduce error and increase efficiency as the clinician reviews the huge quantities of image data.
Once you make sure that the clinician sees as much as possible without extra marking, when they are interested in a specific structure, we provide them with tools so that they can examine that structure exhaustively and accurately, Gupta says. In a series of CT images, an area that originally looks suspect may turn out to be a blood vessel instead. Additionally, CAD serves as a second read to present a nodule that the physician may have missed.
Within Siemens lung package, Lung Care is the portion that can automatically match images from two different exams, and provides the analysis functions. The works-in-progress portion is designed to add the computer-assisted detection function as an added feature to that product. That functionality is currently proceeding through the regulatory process of the FDA.
Philips Medical Systems (Bothell, Wash.) has a commercially available lung nodule assessment product that provides volume measurements with automatic segmenting and facilitates comparison studies for lung nodules that a physician identifies on a CT scan.
Thomas van Elzakker, director of global marketing strategy for CT radiology products for Philips, explains that they have started the PMA process through the FDA for their research prototype for automatic lung nodule detection that is still considered a works-in-progress at this point. They plan to position this product in the market as a second-read support application once they have completed all validation steps.
At Philips, we have seen the need for applications like this, but we also are aware that these applications are there to support and not replace the clinician, says van Elzakker.
CAD for DR
Deus Technologies has entered a strategic partnership with GE Medical Systems (Waukesha, Wis.) to develop a digital CAD product for lung cancer detection for use on GEMS suite of digital radiography (DR) x-ray systems.
Steve Lorenc, general manager of the diagnostic x-ray business, explains that because GEMS has focused heavily in digital radiography (DR) for many years with advances in flat-panel detector technology, the natural evolution of applying CAD to this imaging modality provided a logical progression of capabilities.
Computer-assisted detection applies algorithms in software to detect the presence of nodules in the lung, with a particular focus on early stage lung cancer for nodules in the nine to 15 millimeter range, says Lorenc. These activities are directed primarily towards high-paced screening centers that use the Revolution XR/d or XQ/i DR systems.
With an installed base of 500 worldwide, GEMS considers CAD to be an important enhancement. Still considered a works-in- progress, Lorenc anticipates they will unveil the system at RSNA this fall.
Future for CAD
Deus Freedman asserts that the future of CAD is immense, with applications across all modalities and for numerous disease diagnostic and follow-up procedures. He predicts that CAD will be used for kidney studies and in the liver to automatically measure both metabolic abnormalities that show up as texture and absorption, and for focal lesions.
However, Freedman believes that one of the limiting factors to the development of new CAD applications could be the lack of well-educated computer scientists who are capable of advancing these techniques. He applauds the initiatives from the National Cancer Institute to develop and make available necessary materials to educate computer science and engineering students in CAD.
Vital Images Miller expects that the next most likely CAD initiative will be to use 3D visualization for the detection and assessment of colon polyps. Since many people avoid the gold standard test, colonoscopy, for colon cancer detection, CT colonography may help to eliminate some cancer deaths because it is a less invasive and more acceptable alternative for some patients.
Miller suggests that another interesting opportunity for CAD may involve the detection and assessment of soft plaque in arteries for the management of patients at risk of stroke or cardiac arrest.
We have a CT cardiac package that not only looks at calcified hardened plaque, but we also can see soft vulnerable plaque which is the really bad plaque because it causes heart attacks and strokes that kill or debilitate thousands of people, says Miller. Combined with a CAD capability, the detection of these plaques may be enhanced.
Siemens Gupta raises the possibility of applying CAD techniques to scans produced by hybrid scanners, such as their Biograph that combines the functional data from positron emission tomography with the anatomic detail of CT.
The tools are very similar, says Gupta. You need segmentation, you need to detect, you need to isolate the structures and quantify them, and track them over time to see how the patient is progressing. The paradigm of CAD is now all of that.
Rogers of CADx sees the integration of data from throughout the medical setting to improve the diagnostic process. Because CAD analyzes patterns (and text includes word patterns), clinical information from the hospital information system (HIS) that might include physician notes, EKG tracings and their reports, and patient history could be combined with image data to further enhance the clinicians diagnostic capabilities.
Eventually, what were going to get with our CAD is something we call interactive intelligence, explains Rogers. In that case, not only would the CAD report out, but also the human would be able to query. The response could include images or text, or a combination of integrated data. In this scenario, CAD becomes more than just a post-processing function, but will be integrated within the workflow and interact like another entity that affects the physicians decision-making process.
Because the computer never forgets information that has been entered into its database, it could prompt and remind the physician to consider an additional of possible diagnoses. CADx Kabrisky describes this functionality as a very bright graduate assistant who is there to help the physician.
Computer-assisted detection shows great promise in assisting the physician in both diagnostic and follow-up processes to manage patient care in a variety of clinical settings. Originally developed in mammography, CAD has moved into lung nodule detection and may well move into other clinical areas as the years unfold. The functional capabilities may grow to include data from a variety of sources other than images produced by sophisticated scanners. Patient care becomes the beneficiary of this additional data review.