Software boosts accuracy, reduces reading time for bone scans

November 29, 2006

Software designed by researchers at the University of Chicago helps detect interval changes in successive nuclear medicine bone scans, and can reduce interpretation time by up to 32%, according to research presented in an educational exhibit at the RSNA meeting.

Software designed by researchers at the University of Chicago helps detect interval changes in successive nuclear medicine bone scans, and can reduce interpretation time by up to 32%, according to research presented in an educational exhibit at the RSNA meeting.

Scintigraphic bone scans represent about 25% of all nuclear medicine procedures. Whole-body bone scans have a high sensitivity for detecting bone abnormalities and bone destruction that occurs from primary bone tumors, skeletal metastases, osteomyelitis, and fractures. But for radiologists, the process of identifying and evaluating multiple lesions when comparing successive scans can be time consuming.

And it's not uncommon for radiologists to under diagnose these subtle changes, according to Junji Shiraishi, Ph.D., who led the research team.

"It can be very difficult to compare two different studies with different image characteristics with respect to intensity, position, and size, especially if a large number of bone abnormalities are visible," he said. "The ability to identify changes automatically will reduce reading time for radiologists and may help improve their diagnostic accuracy."

Shiraishi and colleagues developed a computerized temporal subtraction (TS) technique designed to highlight interval changes between successive whole-body scans. Because patients cannot be positioned identically for repetitive whole-body scans, the software uses a nonlinear warping method as part of the TS technique in order to correlate the two images. A nonlinearly warped previous image is subtracted from the current image, and changes are highlighted on the third subtraction image.

In the study, 20 pairs of successive whole-body bone scans were randomly selected from 1038 procedures performed at the University of Chicago in 2004. Selection criteria mandated that each scan include both posterior and anterior views; have at least one abnormal finding in one or both scans of a patient; and demonstrate a maximum of 20 interval changes as determined by a consensus of three radiologists.

In the first of two sessions, five radiologists were shown the pairs of images, which were marked for changes. In the second session held at least two weeks later, the TS image was added to each pair.

With the use of TS images, the average sensitivity in detecting interval changes increased from 58.6% to 73.2%, with a false positive rate of 2 per case. Reading time for each of the five participating radiologists was reduced by 20%, with the mean reading time reduction averaging 32.5%.

The software takes only a few seconds to generate the image, and may be done automatically when successive whole body scan exams are selected at a PACS diagnostic workstation. A newly initiated clinical trial is now under way with the hope of generating the same levels of accuracy and productivity.