Duke researchers hammer away on CAD system to help sift large tomo data sets
While the potential of breast tomosynthesis has received attention in the past year, a commercial chest tomosynthesis scanner will soon be on the market. A computer-aided detection system tailored for the chest application is being fine-tuned, although no date has been set for commercial availability.
James T. Dobbins III, Ph.D., director of the medical physics graduate program at Duke University in Durham, NC, has researched chest tomosynthesis for 20 years. For the past five of those, Dobbins' lab has collaborated with GE Healthcare, making it possible for the vendor to receive FDA approval earlier this year for its VolumeRad system. The device will be commercially available by the end of the year, according to Kristin Binns, global communications manager for GE.
Chest tomosynthesis has practical advantages, Dobbins said. It has better resolution than CT in the plane of the image but worse resolution in the depth direction. For that reason, it is not a replacement for volumetric imaging. Its niche will be as an adjunct to conventional radiography. It is relatively low dose and low cost, and it can be performed on a standard digital chest radiography system at the same time a patient receives a chest film. The scan can be acquired in 10 seconds and requires no patient repositioning.
In an ongoing trial, Dobbins and colleagues are acquiring chest tomosynthesis scans of subjects with known pulmonary nodules scheduled for CT follow-up. Preliminary analysis of 150 patients looks promising in terms of highlighting pulmonary nodules relative to what can be seen in a PA radiograph alone, Dobbins said.
Chest tomosynthesis might also be used to follow up nodules, rather than tracking tumor status with CT, Dobbins said.
Preliminary results using a CAD algorithm developed at Duke are encouraging. CAD detected 11 of 12 nodules, ranging in size from 4 mm to 21 mm, in five patients. The study was presented at the 2005 RSNA meeting.
Use of CAD for evaluating conventional chest radiography suffers from false positives due in part to vessels and rib crossings that can mimic nodules. While tomosynthesis can eliminate these problem areas, it can fall prey to another problem: end-on vessels seen in cross section, which can resemble nodules.
In the 2005 study, CAD marked an average of seven false positives per subject, which was a lower rate than that for conventional radiography but not low enough to satisfy researchers. Investigators developed the original CAD algorithm on conventional chest radiographs and then applied it to tomosynthesis. Their next task is to train the CAD system on tomosynthesis images, with an emphasis on reducing false positives by focusing on vessel cross sections.