CAD boosts performance of rookie CT colonographers

November 29, 2005

CT colonography is not an easy exam to perform and interpret. But intense training and the use of computer-aided detection could improve polyp identification for rookie readers, according to a pair of studies presented Tuesday.

CT colonography is not an easy exam to perform and interpret. But intense training and the use of computer-aided detection could improve polyp identification for rookie readers, according to a pair of studies presented Tuesday.

CAD improves reader sensitivity at the expense of an increase in false positive rates, according to researchers at the Cleveland Clinic Foundation. Dr. Mark Baker, head of abdominal imaging, and colleagues assessed the effect of adding CAD to the interpretations generated by inexperienced readers. The study included seven inexperienced readers from six centers.

Researchers reported that the average sensitivity of polyp detection per patient rose from 0.738 to 0.845 with the addition of CAD, and the overall false positive rate rose from 0.09 to 0.106.

Additionally, CAD added about five minutes to interpretation during the first 10 cases read, Baker said. That number dropped to just two additional minutes for the final 10 cases read, indicated a learning curve for CAD.

"CAD is an essential adjunct to primary CT colonography interpretation, especially in the case of an inexperienced reader," Baker said.

Because there is a steep learning curve for CT colonography interpretation, training is vital, said Dr. Stuart Taylor, of the University College Hospital, London.

Inexperienced readers should receive hands-on training, read a minimum of 40 cases, and learn about tagging methods and 2D and 3D interpretations, according to Taylor. Researchers led by Taylor attempted to determine whether training was still needed when readers were using CAD systems.

Graph demonstrates improvement in CAD-assisted performance of six readers after one day of dedicated CT colonography training. Overall reader sensitivity for polyps of all sizes increased by 18%. (Provided by S. Taylor)

An earlier study conducted by the group found that use of CAD improved sensitivity in inexperienced readers by 39%. For the work presented on Tuesday, investigators pulled six readers from the original study and asked them to read 120 CT colonography cases. Researchers then recorded interpreters' reporting times and diagnostic confidence.

The readers underwent an intensive day of training two months later. After one week, they reread 20 abnormal cases using the CAD system. The dataset included 55 abnormal findings, of which CAD highlighted 55% of the small polyps (1 to 5 mm), 67% of the medium polyps (6 to 9 mm), and 89% of the large polyps (10+ mm).

One day of training was enough to significantly increase polyp detection in four out of the six readers, according to study results. Overall, training increased sensitivity by 18% for all polyp sizes. Readers also reported increased diagnostic confidence.

The flip side to the focused training was an increase in false positives. The median total number of false positives increased significantly, from 2.5 to 5.5. Training also increased mean reading time from 11 to 16.5 minutes.

"CAD alone cannot replace the need for training," Taylor said. "Just one day of focused training significantly improves reader performance and confidence."