The study included 20 CT colonography data sets. Ten were normal and the other 10 included 11 polyps 5-12 mm in size, said Hiro Yoshida, PhD, assistant professor of radiology at the University of Chicago. Four readers (two attending radiologists, a radiology resident and a gastroenterologist) reviewed the data sets without the CAD system; they reviewed them again with CAD, Dr. Yoshida said. The CAD system shows physicians the possible location of suspicious polyps in an effort to significantly increase their detection performance, he said.
Human observers missed 42% of the polyps without CAD; 75% of the missed polyps were later identified by the human observers with the help of the CAD system, Dr. Yoshida said. The CAD system was actually able to detect all of the polyps missed by the readers; however, the readers "disagreed" with the computer's assessment. Dr. Yoshida noted that the low 42% figure for human observers is due in part to the fact that "we selected very difficult cases to read and we only provided the reader with one view of the colon."
There are areas where the computer is wrong, said Dr. Yoshida. "However, our study found that 77% of the computer's false positives were easily (and appropriately) dismissed by the human readers," he said.
"It is very difficult and time consuming to read and interpret CT colonography examinations," said Dr. Yoshida. The time and the difficulty of the doing and interpreting the examination could be a barrier to more widespread colon cancer screening. "CAD has the potential to play an important role in massive screening," he said.
The CAD system is currently being developed at the University of Chicago. It is not widely available at this time. However, it has been licensed to two companies, R2 Technology, Inc and Median Technologies, and
Contact: Keri Sperry
American Roentgen Ray Society