Defining Lung CADs Role

Twitter icon
Facebook icon
LinkedIn icon
e-mail icon
Google icon
  The right and left images are from the same patient but acquired at different dates. In assessing pulmonary nodule evolution by automatic computation of volume and diameter changes, Median Technologies’ LMS-Lung CAD automatically retrieves the lesion in the new dataset, measures it, providing variation of volume and diameters.

Computer-assisted detection (CAD) software to analyze lung imaging studies is steadily gaining ground, improving in sensitivity and specificity as a increasingly reliable second pair of trained eyes to aid radiologists in the detection of pulmonary nodules. While more radiologists are championing the technology, the lack of PACS integration is a stumbling block for widespread adoption. At the same time, what first emerged as a tool for detection is transforming into a tool for characterization and possible diagnosis, while expanding its utilization beyond pulmonary diseases and lung cancer.

A second reader that makes sense

Lung CAD software was developed primarily for the detection of pulmonary nodules, which can be overlooked by radiologists reading chest x-rays and lung CTs. The software has been refined over the last few years, overcoming initial speculation regarding its efficacy with recent clinical studies that have validated its performance.

“Chest x-ray CAD makes sense—it helps to find nodules in the lungs on an x-ray that has superimposed soft tissues, skeletal structures, mediastinal structures—the problem is that it seems to be oversensitive in areas of overlapping bones such as the anterior rib and posterior rib crossings,” says Judith K. Amorosa, MD, clinical professor, radiology at Robert Wood Johnson School of Medicine in Piscataway, N.J., who uses the IQQA-Chest CAD enterprise solution from EDDA Technologies.

Lung CT CAD finds nodules which need to be distinguished from blood vessels and pleural reflections. It also can be useful for less experienced readers for confirmation of a possible finding. Used by less-experienced readers, CAD for chest x-rays (CXRs) will likely lead to more chest CTs, however, it is less expensive than a second reader, Amorosa notes.

Achieving a high level of sensitivity and specificity with the actual software, at a level where radiologists find it helpful in their day-to-day work, has been a challenge, adds Heber MacMahon, MB, BCh, professor of radiology at the University of Chicago and director, section of thoracic imaging.

Using the OnGuard lung CAD system from Riverain Technologies, MacMahon and colleagues sought to determine the sensitivity and number of false-positive marks for identifying lung cancers that were detected but misinterpreted on chest radiographs by radiologists. The OnGuard system marked 35 percent of lung cancers that had been previously overlooked, according to the study results that appeared in Radiology (January 2008).

Despite its sensitivity, the software tends to have a relatively high number of false positives, marking not only the nodules but also parts of normal anatomy which becomes distracting and time-consuming for the radiologist, he says, stressing the importance of reducing the number of false positives while maintaining sensitivity so tool aids radiologists rather than slowing them down.

Another bone of contention is determining not when the software should be applied, but how. Lung CAD might have started as a means of detection but systems are now being developed to diagnose and characterize lung nodules, offering temporal comparisons to give growth rate and doubling time. Some systems also offer tools that address density, borders and the overall appearances of the nodule, says Heidi Roberts, associate professor of radiology, University Health Network in Toronto.

“The benefit to temporal characterization that will quantify these features is ultimately an improvement in workflow and patient management,” Roberts adds. For example, in a patient with interstitial lung disease, lung CAD can be helpful in characterizing the disease, with the radiologist and pulmonologist being able to assess whether the patient is responding to treatment.

“Every CAD software has its strengths and weaknesses—the question that is difficult to answer is, do we only need CAD for lung nodule detection or do we want it for additional disease detection?” she notes, adding that lung CAD software vendors are looking outside nodule detection to other pulmonary diseases such as emphysema quantification or other airway diseases. Further studies,