With early diagnosis the only way to beat lung cancer, lung CAD is focused on finding tumors of 2 to 8 millimeters or more imaged via chest CT or analog or digital x-ray. While it’s too early to say if lung CAD will be successful in bringing better outcomes for lung cancer patients — it is beginning to make its mark.
The challenges of detecting lung cancer at a very early stage are formidable, but the urgency propels the development of tools for early detection. The prognosis for surviving beyond five years is abysmal — 9 to 12 percent — for patients who are diagnosed with more advanced stages of this disease. The National Cancer Institute released a report in August 2005 detailing that 160,000 people in the United States die from lung cancer each year confirming it again as the No. 1 cause of cancer deaths in this country.
Lung CAD (computer assisted detection) software has been developed for electronic analysis of both analog and digital chest radiographs and to scrutinize the large (and growing!) data sets created by multidetector CT scans. This tool is employed as a secondary read where the radiologist accepts or rejects CAD’s marked areas of interest after a primary read.
Reimbursement is critical to the adoption of new imaging technology, so it bodes well that less than a year after the 2004 Radiological Society of North America (RSNA) meeting where Riverain Medical launched its Rapid-Screen chest x-ray CAD system, the American Medical Asso-ci-a-tion issued a Category III CPT code to cover chest x-ray CAD that became effective on Jan. 1, 2006. Lung CAD for CT exams is not yet reimbursed.
Heber MacMahon, MD, a pioneer and leader in the development of lung CAD, professor of radiology at the University of Chicago and consultant to Riverain Medical, explains that even the most conscientious and careful radiologist can overlook suspicious nodules. Fatigue is one factor as these clinicians are expected to read more studies in a shorter period of time and in the case of CT, the exponential increase in the quantity of images produced by the multidetector scanners.
Lung CAD and the ubiquitous chest x-ray
Chest x-rays are the most requested examination ordered in a hospital because they are required not only for symptomatic patients, such as those with suspected pneumonia or heart disease, but also those patients considered asymptomatic such as those scheduled for surgical procedures other than those involving the heart and lungs.
MacMahon notes that while chest x-rays are not currently recommended for lung cancer screening, many malignant nodules are picked up incidentally in routine chest x-rays. “In other words, the patient is completely asymptomatic from the cancer, and he or she has a chest x-ray for another reason. If we can detect a questionable lesion on x-ray, that may help.”
He explains that while CT scans are capable of detecting nodules that are 2 millimeters in size, even with CAD under ideal conditions, lung nodules of 5 millimeters push the limits of detection on a chest x-ray. On the other hand, he notes that a very high proportion of nodules that are less than 5 mm are benign, and the more important nodules are 8 mm and larger. Additionally, he anticipates that the accuracy of lung CAD will improve in sensitivity while decreasing the number of false positives that are presented. A false positive might include marking normal anatomy, focal scars or other abnormalities, and they prove to be a distraction for the radiologist and may slow reading times.
Greig Huggins, vice president of business development for the Utah Imaging Associates in Salt Lake City, says their 33 radiologists who cover six hospitals plus, selected the Riverain RapidScreen chest x-ray CAD system to use in analyzing virtually every chest x-ray they review. The system identifies areas of interest that are suspected nodule sites for both analog and digital chest images. In their clinical trials for pre market approval, Riverain demonstrated a 23 percent increase in radiologists’ detection of solitary pulmonary nodules for early stage lung cancer (9-14 mm) when RapidScreen was used. The installed base now numbers about 200.
The RapidScreen system resides on a central server connected to PACS in their basic six hospital network, but is waiting for the hospital protocols to be completed before full deployment of the automated processes. Workflow includes routing DICOM chest x-ray images into the system, and following processing activities, returning the images to their originating source with the basic chest image plus the one that is marked with suspicious areas.
“From an IT perspective, this was very easy with our PACS. We configured the routing rules on our NovaPACS,” explains Huggins. Besides their six primary hospitals throughout the Salt Lake City area, Utah Imaging Associates also provides nighttime, specialized and vacation relief coverage for several small hospitals of 50 beds or less. For those institutions without digital radiography (DR) or computed radiography (CR) acquisition capabilities, they use digitized analog imaging studies for their electronic reads.
Edwin J.R. van Beek, MD, PhD, FRCR, professor of radiology at the Carver College of Medicine at the University of Iowa Hospitals and Clinics selected the IQQA-Chest from EDDA (Embedded Diagnostic Decision Aid) Technology as a real-time interactive diagnostic analysis system for their digital chest x-rays. The system received FDA clearance in October 2004. He sees several roles for CAD including early detection.
“We started using it with patients with known cancer whom we were following up to look for metastases with chest x-ray, such as breast cancer or prostate [cancer],” van Beek explains. “For those patients, it has been quite successful in offering us a second set of eyes.” In addition, when used as a detection capability, he believes this offers a better utilization for CT when they observe a suspicious lesion on chest x-ray.
They have their current system configured on a separate server, but that is just because they are setting up a new PACS. Once the PACS is live, he sees no reason that they could not set up the software within the PACS. Of course, he notes that there is a convenience to having the IQQA-Chest on a separate server because it generates reports which can be saved for comparison later. They are able to save the report and the key images that demonstrate the size of the nodule, whether or not it is calcified, and whether they believe it to be benign or malignant for future comparison.
“A relevant factor involving the detection rate of nodules is crucially dependent on how the algorithm is built for the system,” van Beek concludes. “We chose this system [which they helped to develop through testing] because it is focused on 5 to 15 mm nodules.”
Multi-center clinical studies suggest that using IQQA-Chest interactively within the clinical routine, the detection of actionable small nodules could be increased to over 80 percent. The current installed base includes 10 sites worldwide.
CT and lung CAD are powerful allies
The use of lung CAD coupled with CT offers the benefit of visualizing smaller nodules and some lesions that might otherwise go completely undetected. However the expense and radiation dose of CT raise cautionary flags to using this technology for routine screening.
Additionally, while almost all lung cancers begin as pulmonary nodules, most of the nodules prevalent on CT scans are not cancerous. The challenges of designing a CAD system that is capable of identifying nodules accurately, eliminating as many false positives as possible, while providing information about whether or not those lesions are malignant serve as foundational issues in lung CAD.
Philippe Grenier, MD, professor of radiology at the Université Pierre- et-Marie-Curie in Paris, has been using the Median Technologies Lung CAD (Median CAD-Lung) product for about a year to follow and evaluate growth rate for detected lung nodules.
This system is capable of quantification of volumetric changes of a given nodule from one scan to another. “Because nodules are a sphere, when the volume doubles, the diameter only increases by 25 percent,” he explains. “So a 5 mm nodule that doubles in volume, the diameter increases only by 1.25 mm to 6.25 mm.”
Using this system to compare successive CT scans, they can accomplish a 3D reconstruction with automatic measurements that are robust and reproducible. “A change in volume that is greater than 20 percent is regarded as significant.” This software automatically registers the images produced in successive scans.
Besides using this CAD program for detection of solid nodules, Grenier appreciates its ability to depict non-solid lung lesions that have a low density and are much more difficult to observe even though they have a greater probability of being malignant than solid nodules. “This system is very good at depicting the ground glass opacity non-solid nodules.”
In one study, Median Technologies CAD-Lung software used as a second reader in routine reporting nearly doubled the percentage of patients with actionable nodules larger than 4 mm. The company anticipates this product, which received a CE mark in Europe, to have about half a dozen installations over the next quarter. Although it has not yet received clearance from the FDA, it is being used at three U.S. research sites.
David S. Mendelson, MD, associate professor of radiology at Mount Sinai School of Medicine in New York City, has utilized the R2 Technology ImageChecker CT as a clinical test site. They found that not only was this system quite good at detecting nodules — especially in the 4 mm to 1 cm range (optimized for 4 to 8 mm) — but what they found most impressive was the user interface that is designed to streamline workflow.
Mendelson’s workflow involves reading the exam first without CAD markings, looking for nodules. Then he reviews the R2 results. If it confirms one of his findings, he moves on. If it detects something he did not see, he decides whether or not it is a “real” concern and disposes of those that are false positives, an action that he finds easy to do.
“My general opinion is that CAD is not a time saver, rather it improves quality, and to me the point is to be able to improve quality quickly and efficiently,” he says.
Two other features of the system he appreciates are the ability to send screen shots of the CAD capture into the PACS so that images can be reviewed at any workstation on the network, as well as a temporal comparison feature that compares nodules from one scan to another. Currently this must be performed on the R2 workstation but integration with other vendors’ products is in the works.
In their clinical research submitted to the FDA in 2004, R2 demonstrated that the ImageChecker CT reduced the number of actionable lung nodules missed by radiologists by 26 percent, with a false marker rate of only two per normal case. Thirty-three 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 that were detected by CAD. Some 100 units have been installed to date.
Arnold C. Friedman, MD, FACR, who is a professor of radiology at the University of Florida in Jacksonville, has been evaluating the iCAD CT Lung product for screening and instances in which a CT scan has been done for another purpose as part of the FDA approval process. This system continues to undergo FDA review at this time.
Just as with mammography screening, when the vast number of imaging studies are normal, it is easy to miss nodules, especially given the huge image data sets produced in a chest study by multidetector CT scanners. “Having a computer flag the nodules will decrease the number that are missed,” Friedman says.
Besides use in a dedicated screening program, he notes that sometimes nodules can be missed when the radiologist is not specifically looking for them. For example, when a CT of the abdomen is done, and a portion of the lung is captured on the scan, the CAD would mark any nodules that otherwise might be missed.
University Health Network in Toronto has been a beta test site for Medicsight LungCAD software system for reviewing their CT scans during the past year. They have used this system for research purposes only, not for clinical management of patients.
The study involves early detection of lung cancer using low-dose CT in individuals who are current or former smokers, and their current enrollment in the study stands at almost 1,100 subjects. Heidi Roberts, MD, associate professor of Radiology in the Department of Medical Imaging, has found this CAD system has potential to improve their workflow, especially given large CT studies with thinner and thinner slices. In their department, they may do 30 to 50 studies a day with 250 or more slices per scan.
Considering that all lung CAD programs produce a number of false positive readings, Roberts says that with this system those are easy to dismiss. She appreciates the filters they can set to show only nodules of a certain size as well as its shape, such as round or tubular. This program can be customized to each user as well, adjusting for more or fewer marks per image depending on user preference.
Besides screening, she believes CAD will be important to managing oncology patients when every nodule is important because they often represent metastasis. Future questions she raises include the evaluation of CAD performance with CTs obtained with different radiation doses and to assess how CAD performs at different slice thicknesses in reconstruction.
Medicsight LungCAD adopts a different paradigm for application as a concurrent-read model where sensitivity can be controlled by the reading radiologist. This software has been packaged as an API (application program interface) algorithm for use on both Viatronix and TeraRecon workstations. Clinical research sites include medical centers in England, Spain and Canada.
With mammography CAD having achieved a high level of success, many clinicians look to lung CAD to assist in similar ways and with comparable results. At current, lung CAD systems are still evolving with an eye toward improving sensitivity while reducing the number of false positive findings.