Computer-assisted detection of pulmonary nodules:: evaluation of diagnostic performance using an expert knowledge-based detection system with variable reconstruction slice thickness settings

The purpose of this study was to evaluate the performance of a computer-assisted diagnostic (CAD) toot using various reconstruction slice thicknesses (RST). linage data of 20 patients undergoing multislice CT for pulmonary inetastasis were reconstructed at 4.0, 2.0 and 0.75 rnrn RST and assessed by two blinded radiologists (R1 and R2) and CAD. Data were compared against an independent reference standard. Nodule subgroups (diameter >101, 4-10, <4 rnm) were assessed separately. Statistical methods were the ROC analysis and Mann-Whitney U test. CAD was outperformed by readers at 4.0 mm (Az = 0.18, 0.62 and 0.69 for CAD, R I and R2, respectively; P<0.05), comparable at 2.0 mm (Az = 0.57, 0.70 and 0.69 for CAD, RI and R2, respectively), and superior using 0.75 min RST (Az = 0.80, 0.70 and 0.70 and sensitivity = 0.74, 0.53 and 0.53 for CAD, RI and R2, respectively; P<0.05). Reader performances were significantly enhanced by CAD (Az = 0.93 and 0.95 for R1 + CAD and R2 + CAD, respectively, P<0.05). The CAD advantage was best for nodules <10 min (detection rates 93.3, 89.9, 47.9 and 47.9% for RI + CAD, R2 + CAD, RI and R2, respectively). CAD using 0.75 min RST outperformed radiologists in nodules below 10 mm in diameter and should be used to replace a second radiologist. CAD is not recommended for 4.0 mm RST.