Assessment of an optical flow field-based polyp detector for CT colonography
Most current computer-aided detection (CAD) algorithms for the fully automatic detection of colonic polyps from 3D CT data suffer from high false positive rates. We developed and evaluated a post-processing algorithm to decrease the false positive rate of such a method. Our method attempts to model the way a radiologist recognizes a polyp while scrolling a cross-sectional plane through 3D CT data by quantifying the change in location of the edges in 2D plane. It uses a classifier for identification based on the Mahalanobis distance. The new method increased the ROC curve area from 0.89 to 0.98 (an increase from 34.5% to 85.0% in specificity for 100% sensitivity) in a population of 8 patients.