Computer-generated colonic polyps can serve as a gold standard for evaluation of methods to visualize and detect precancerous lesions in the colon, Stanford researchers reported at an RSNA scientific session.
Digitally placed within volumetric spiral CT images of the colon, the synthetic polyps proved indistinguishable from real polyps when evaluated by radiologists blinded to the nature of the images.
"The radiologists were shown images with an arrow pointing to a polyp, and they were asked whether the polyp was real or synthetic," said Sandy Napel, Ph.D. a radiology physics investigator at Stanford. "They could not tell the difference between real and synthetic polyps."
The polyps are developed by use of CT simulator software configured to mimic the geometry and scanning protocol of a spiral CT scanner. Additional software was developed to permit insertion of polyps into CT scans, Napel said.
The research is part of an ongoing evaluation of CT-based virtual colonoscopy as an alternative to fiber optic colonoscopy (FOC), the current gold standard for colon evaluation. Napel said that FOC has a number of limitations, such as operator-dependent accuracy and inability to visualize the entire colon.
The synthetic polyps were compared against colonic images showing actual polyps. Three radiologists blinded to the nature of each image were asked whether the polyp depicted in the image was real or synthetic. The images were obtained from eight patients who had FOC that identified at least one polyp.
The evaluation involved 10 real and 10 synthetic polyps, which were depicted in two-dimensional axial and three-dimensional views. For either 2-D axial or 3-D images, the radiologists showed no significant difference in their ability to distinguish real polyps from synthetic ones or vice versa, Kapel said.
"We think this method can be used to generate gold-standard data for the comparison of visualization and detection methods," he said.
