RATIONALE AND OBJECTIVES: To assess the performance of a computer-aided detection (CAD) algorithm for measuring polyp-like structures on CT colonography (CTC) images of a phantom. MATERIALS AND METHODS: We constructed a Plexiglas phantom to which we affixed a series of idealized Plexiglas polyp-like objects, including spheres and hemispheres. We imaged the phantom in a four-channel detector CT scanner at a 1.3 mm slice thickness with a reconstruction interval of 0.6 mm, using combinations of 100 mAs, 30 mAs, horizontal and vertical orientation. For each set of CT images, the interior surface of the phantom was segmented. The CAD algorithm was applied to the resulting surface to identify the polypoid regions of interest and to calculate their volume and maximum linear dimension. Calculated values were then compared with actual values to yield percent error in each measurement. RESULTS: The mean error in volume for the subgroups of spheres and hemispheres was 3% and 5% respectively. Mean error in linear dimension was approximately 2% for both shape subgroups. All CAD-calculated values were closely correlated with their respective actual values. Parameter selection did not significantly affect the accuracy of the calculated measurements. CONCLUSIONS: Our CAD software accurately measured the greatest linear dimension and the volume of each of the polyp-like structures in our phantom. Results were largely independent of phantom orientation and the CT exposure factors.
RATIONALE AND OBJECTIVES: To assess the performance of a computer-aided detection (CAD) algorithm for measuring polyp-like structures on CT colonography (CTC) images of a phantom. MATERIALS AND METHODS: We constructed a Plexiglas phantom to which we affixed a series of idealized Plexiglas polyp-like objects, including spheres and hemispheres. We imaged the phantom in a four-channel detector CT scanner at a 1.3 mm slice thickness with a reconstruction interval of 0.6 mm, using combinations of 100 mAs, 30 mAs, horizontal and vertical orientation. For each set of CT images, the interior surface of the phantom was segmented. The CAD algorithm was applied to the resulting surface to identify the polypoid regions of interest and to calculate their volume and maximum linear dimension. Calculated values were then compared with actual values to yield percent error in each measurement. RESULTS: The mean error in volume for the subgroups of spheres and hemispheres was 3% and 5% respectively. Mean error in linear dimension was approximately 2% for both shape subgroups. All CAD-calculated values were closely correlated with their respective actual values. Parameter selection did not significantly affect the accuracy of the calculated measurements. CONCLUSIONS: Our CAD software accurately measured the greatest linear dimension and the volume of each of the polyp-like structures in our phantom. Results were largely independent of phantom orientation and the CT exposure factors.