Computer-aided detection of polyps in CT colonography:: Evaluation of volumetric features in differentiating polyps from false positives

We evaluated the ROC performance of six volumetric features in differentiating between the true-positive and false-positive polyp findings of our CAD system for CT colonography. Among the single feature statistics, the mean shape index had the best ROC performance, followed by a new D-iris feature. The ROC results of the round-robin discriminant analysis indicate that the performance of our CAD system can potentially be improved by use of a combination of the shape index, CT value, and the D-iris feature statistics.