Jiamin Liu PhD, Presenter: Nothing to Disclose

Yucherng Chang BS, Abstract Co-Author: Nothing to Disclose

Ronald M. Summers MD, PhD, Abstract Co-Author: Royalties, iCAD, Inc, Nashua, NH Grant, iCAD, Inc, Nashua, NH Stockholder, Johnson & Johnson

Oral contrast agents often tag polyp surfaces in a pattern that is distinct from internal tagging of residual stool. The aim of this study was to develop a method for automated detection of adherence of contrast agents on polyps in CT colonography.

A method for automated classification of the relationship of polyps to intraluminal oral contrast for CTC was developed. The polyps were classified into one of three categories: Adherent Contrast, No Contrast, or Contact/In contrast agent pool. For each polyp, first, a VOI (volume of interest) was automatically determined around the segmented polyp. Then, presence of contrast agents within the VOI was computed by thresholding. A polyp within a VOI that contained oral contrast was pre-classified as Adherent Contrast or Contact/In contrast agent pool. A polyp within a VOI that did not contain oral contrast was pre-classified as No Contrast or Adherent Contrast. Finally, morphologic operation (dilation/erosion) and connectivity property are utilized to refine those pre-classified polyps for final classification. Performance of the automated classification is evaluated using 589 optical colonoscopy-proven polyps identified on CTC of 212 patients with polyps in a larger screening population. Sensitivity and specificity with 95% confidence intervals were reported.

The classification sensitivities for No Contrast, Adherent Contrast, and Contact/In agent pool were 89.9% (257/286; [0.86, 0.93]), 77.8% (133/171; [0.71, 0.84]), and 99.2% (131/132; [0.95, 1]), respectively. Their specificities were 87.5% (265/303; [0.83, 0.91]), 94.7% (396/418; [0.92, 0.97]), and 98.2% (449/457; [0.96, 0.99]), respectively. The reason for the lower sensitivity (77.8%) of Adherent Contrast is mainly due to the partial volume effect when the agent slightly coats the polyps and those polyps were misclassified as No Contrast. For polyps with villous histology, sensitivity of Adherent Contrast was 95.5% (21/22; [0.75, 1]).

The automated classification of adherent contrast agents on polyps achieved high sensitivity and specificity. The method was highly accurate for the villous polyps that convey the highest risk.

A CAD system with this technique may improve the accuracy of colorectal cancer screening.

Liu, J, Chang, Y, Summers, R, Automated Detection of Adherent Contrast Phenomenon on Colonic Polyps on CT Colonography.  Radiological Society of North America 2009 Scientific Assembly and Annual Meeting, November 29 - December 4, 2009 ,Chicago IL. http://archive.rsna.org/2009/8005678.html