Lung Cancer: CAD Provides a Closer Look

Twitter icon
Facebook icon
LinkedIn icon
e-mail icon
Google icon
LungCAD from Medicsight helps radiologists detect small lung nodules as well as better measure growth to determine changes that indicate a malignancy.

Applying CAD to chest and lung imaging has a special urgency, since the five-year lung cancer survival rate is so low and early detection is so difficult. Imaging advances are honing in on smaller and smaller nodules and research shows that CAD is helping radiologists find more of those nodules. But for lung CAD to truly gain believers, clinical validation studies are needed as well as smooth integration into radiology workflow.

Mary Roddie, MD, a thoracic radiologist at Charing Cross Hospital in London, began working with Medicsight LungCAD during a sabbatical that coincided with the product’s development period. Because of her interest in improving the ability to measure lung nodule volume, she participated in beta testing and has been using Medicsight’s product experimentally in her practice.

One challenge arises from lung cancer screening, she says. “Nearly everybody has nodules in their chest and the vast majority are benign. The goal is to see whether the nodules are growing over time. To do that accurately, we need to be able to make accurate measurements.”


Advancing accuracy



The trouble is, measuring very small nodules accurately is difficult. Nodules more than 10 millimeters (mm) in diameter are easily measured by most software so “those are not really clinical problems for us,” Roddie says. “Once a nodule is 1 centimeter (cm) or more, we’re taking action because there is a high likelihood of its being malignant.” Nodules between 5 and 9 mm in diameter are challenging. A German study conducted in 2003 found that when patients were scanned twice just a few minutes apart, measuring software found an error rate of up to 20 percent between the two images. So, for small nodules, you have to be sure that volume increased more than 20 percent before taking action because anything less could be just an error of measurement. Work to reduce that 20 percent error rate is ongoing, says Roddie.

The other primary use of CAD in lung imaging is for detecting nodules that the radiologist might have overlooked. “I think that is useful, particularly in terms of saving time for radiologists when they’re looking at cancer patients in follow up,” she says. “It’s quite easy to overlook a small metastasis which could have big implications for the patient.”

CAD’s ability to improve radiologist performance has been well documented, says Roddie. “The number of nodules that a radiologist picks up will be improved by CAD. Applied to the right population, that will be a good thing.” While skeptics say it’s yet to be proven whether CAD picks up significant nodules, Roddie says that of greater concern is that CAD won’t be routinely applied to chest imaging until it is seamlessly integrated into radiology workflow. Most radiologists work at a PACS workstation, she says. If they have to push a lot of buttons or switch workstations, few will use the tool. “It’s important that CAD is sitting right there and we can just switch it on. That will be the big barrier.”

CAD vendors are working on integration, she says. A lot of the PACS companies are talking to CAD companies because they’re looking for software to add on to really improve radiology workflow. Until it’s widely available, people won’t start using CAD, she says. However, when you have CAD sitting on your workstation and you choose not to use it and you miss something, it’s going to be difficult in court of law to answer why. “We’re all a little worried about being sued, but if we can use CAD without it affecting our workflow, we will.”

Putting a financial value on CAD is difficult, she says, because while it can make radiologists work more efficiently, it also increases the time of individual reads. “It probably won’t make us read more cases, but will probably make us more accurate. It’s clear from the research that CAD and radiologists pick up different nodules.” Because of that, use of both will provide the best sensitivity and accuracy for nodule detection.


Easy integration and use


William Moore, MD, a radiologist at Stony Brook University Hospital in New York, has been using IQQA Chest Enterprise CAD since February. The hospital chose the solution from EDDA Technology because physicians saw a demonstration of the software, then shopped around and decided they liked this product the best. 

The tool is helpful for lung nodule detection and “anything to increase the detection of lung nodules is a good thing,” he says. When he had other physicians at Stony Brook try it, one said, “We have to have this,” Moore reports. Another, older and less computer savvy physician had no trouble using it. “It helps us do the best job we can do.” When considering the return on investment of CAD for lung imaging, Moore says you can either think about making money or helping patients. “We can help patients more than we could ever make money. That makes it an easy decision to integrate CAD.”

The software loads automatically and it takes just two clicks to get CAD results on an x-ray. That adds about 5 to 6 seconds per exam, a very manageable increase. “There was very little growing pain,” Moore says.

Aside from the improvements in patient care, computer technology is a major thrust of the radiology department at Stony Brook, Moore says. Now that they’ve been using CAD for several months, they are “trying to get a handle on how much it increases accuracy. We’re not saying that every nodule CAD finds is a positive finding.”

The tool probably offers the greatest benefit for general radiologists, Moore says. “Chest radiologists already are pretty good at finding nodules, but CAD will make us better.”

David Pandit, administrative director of radiology at Medical City, Dallas Hospital in Texas, uses RapidScreen, the x-ray-based CAD system from Riverain Medical. Users digitize a chest x-ray and apply a CAD algorithm to note suspicious nodules. RapidScreen is strictly a review tool so no additional radiation dose is required. It is compatible with all modalities that generate DICOM 3.0 conformant digital radiography, computed radiography and digitized images.

Pandit’s facility implemented RapidScreen in January after a couple of years of product research. The hospital has been careful about not overutilizing the tool, with strict guidelines about which x-rays warrant its use.

A challenge in deploying RapidScreen, Pandit says, is that it is a new tool so it needs to be marketed to the medical community. However, “the radiologists really like it,” he reports. And, “administration definitely feels that the investment has been worthwhile.”


On the horizon


CAD for chest and lung imaging may venture into detecting emboli, says Roddie. So far, test results have been disappointing because of a high false-positive rate. Other researchers are looking into using CAD to quantify the amount of emphysema patients have in their chest.

For measurement and detection of lung nodules, “we mainly need to reduce the false positives,” she says. Efforts are underway to try to tie CAD in with other modalities to use more computer intelligence to decide on the nature of a nodule, Roddie says.