Abstract
OBJECTIVE: To assess the sensitivity of detection and accuracy of volumetry by manual and semi-automated quantification of artificial pulmonary nodules in an anthropomorphic thoracic phantom on low-dose CT. METHODS: Fifteen artificial spherical nodules (diameter 3, 5, 8, 10 and 12 mm; CT densities -800, -630 and +100 HU) were randomly placed inside an anthropomorphic thoracic phantom. The phantom was examined on 16- and 64-row multidetector CT with a low-dose protocol. Two independent blinded observers screened for pulmonary nodules. Nodule diameter was measured manually, and volume calculated. For solid nodules (+100 HU), diameter and volume were also evaluated by semi-automated software. Differences in observed volumes between the manual and semi-automated method were evaluated by a t-test. RESULTS: Sensitivity was 100 % for all nodules of >5 mm and larger, 60-80 % for solid and 0-20 % for non-solid 3-mm nodules. No false-positive nodules but high inter-observer reliability and inter-technique correlation were found. Volume was underestimated manually by 24.1 ± 14.0 % for nodules of any density, and 26.4 ± 15.5 % for solid nodules, compared with 7.6 ± 8.5 % (P < 0.01) semi-automatically. CONCLUSION: In an anthropomorphic phantom study, the sensitivity of detection is 100 % for nodules of >5 mm in diameter. Semi-automated volumetry yielded more accurate nodule volumes than manual measurements.