CAD products track changes in lung nodules

The most critical finding in scanning lung cancer patients is the degree of change in suspicious anatomy over time. But while computer-aided detection systems have identified isolated lung nodules, until now they have not clearly separated or segmented them from normal tissue when the lesions are adjacent to blood vessels or the chest wall. Nor have they automated the process of tracking lesions over time.

Developments in products scheduled to hit the RNSA exhibit floor offer a variety of solutions for these problems.

The Temporal Comparison software module from R2 Technology finds the exact location of a nodule in separate CT scans. The product, which passed FDA review in June, also automatically tracks the progression or regression of lung nodules.

"Most radiologists still use calipers or hand-drawn circles around a lung nodule, which is not a repeatable way to measure that object," said Dr. Susan Wood, vice president of CT Products at R2. "What CAD has done is to automate the process to provide a more accurate measure of the change in nodule size."

Temporal Comparison software will debut as a commercial product at the RSNA meeting, alongside R2's ImageChecker CT CAD, the first FDA-approved product for the detection of solid lung nodules during review of multislice CT chest exams. Other similar technologies will be on display as well, but some are far from commercialization.

Deus Technologies has a number of products in the works. One involves image matching, which helps radiologists choose images that most accurately compare CT results obtained at different times. Deus algorithms in development apply various criteria to assess lung regions rich with blood vessels and areas near the periphery. Another product performs temporal subtraction, which illuminates varying rates of change in nodules and blood vessels. These are slated for submission to the FDA in 2005, according to the company.

Lung CAR (computer-assisted reader) software, which received FDA approval in August for CAD newcomer Medicsight of London, U.K., helps identify lung nodules, automatically detecting the outline of a nodule and allowing the radiologist to sharpen nodule boundaries in both 2D and 3D. Lung CAR also automatically extracts and calculates nodule characteristics such as the location and distance to the edge of the lung and the chest wall, the maximum diameter in 3D space, the longest axial dimension, and volume.

Lung CAR will not be displayed by Medicsight at the RSNA meeting, but it will be the subject of a scientific study presented by radiologists from London Chest Hospital. In the study, Dr. Steve Ellis, a consultant chest radiologist at Royal London Hospital, assessed the ability of advanced lung imaging software available on standard CT scanners and Lung CAR to delineate lung nodules using a phantom. He found that advanced lung imaging software had difficulty defining nodules located next to or surrounded by blood vessels and, therefore, could not be used to calculate present and subsequent volumes of the lesions themselves.

"Clinically, when I put standard lung imaging software on a nodule close to the chest wall or a blood vessel, I can't use the data for subsequent follow-up of that nodule because I don't know what proportion of the vessel or the chest wall will be included next time," he said.

Medicsight Lung CAR not only improved the differentiation of nodules from the chest wall and blood vessels; it also allowed users to adjust the boundaries of nodules in a semiautomatic way by adjusting the sensitivity of boundary detection, Ellis said.

"I'm more interested in being able to extract a nodule accurately, so I have a volume I can use to work out whether it's growing. At the moment, clinically, probably 50% of nodules cannot be extracted with advanced lung imaging software because they are adjacent to vessels or the chest wall. So any nodule extraction software that gives me a volume is a boon," he said.

Other scientific papers to be presented at the RSNA meeting will demonstrate that ImageChecker CT CAD yields about a 20% reduction in the number of missed nodules and that CAD performance in detecting lung nodules 4 mm or larger is stable across variations in MSCT collimation, dose, and reconstruction, Wood said.