Model-Based Iterative Reconstruction Technique for Ultralow-Dose Computed Tomography of the Lung

OBJECTIVES: The aim of this study was to assess the effectiveness of a model-based iterative reconstruction (MBIR) in improving image quality and diagnostic performance of ultralow-dose computed tomography (ULDCT) of the lung. MATERIALS AND METHODS: The institutional review board approved this study, and all patients provided written informed consent. Fifty-two patients underwent low-dose computed tomography (LDCT) (screening-dose, 50 mAs) and ULDCT (4 mAs) of the lung simultaneously. The LDCT images were reconstructed with filtered back projection (LDCT-FBP images) and ULDCT images were reconstructed with both MBIR (ULDCT-MBIR images) and FBP (ULDCT-FBP images). On all the 156 image series, objective image noise was measured in the thoracic aorta, and 2 blinded radiologists independently assessed subjective image quality. Another 2 blinded radiologists independently evaluated the ULDCT-MBIR and ULDCT-FBP images for the presence of noncalcified and calcified pulmonary nodules; LDCT-FBP images served as the reference. Paired t test, Wilcoxon signed rank sum test, and free-response receiver-operating characteristic analysis were used for statistical analysis of the data. RESULTS: Compared with LDCT-FBP and ULDCT-FBP, ULDCT-MBIR had significantly reduced objective noise (both P <; 0.001). Subjective noise on the ULDCT-MBIR images was comparable with that on the LDCT-FBP images but lower than that on the ULDCT-FBP images (P <; 0.001). Artifacts on ULDCT-MBIR images were more numerous than those on the LDCT-FBP images (P = 0.007) but fewer than those on the ULDCT-FBP images (P <; 0.001). Compared with the LDCT-FBP images, ULDCT-MBIR and ULDCT-FBP images showed reduced image sharpness (both P <; 0.001). All the ULDCT-MBIR images showed a blotchy pixelated appearance; however, the performance of ULDCT-MBIR was significantly superior to that of ULDCT-FBP for the detection of noncalcified pulmonary nodules (P = 0.002). The average true-positive fractions for significantly sized noncalcified nodules (≥4 mm) and small noncalcified nodules (<;4 mm) on the ULDCT-MBIR images were 0.944 and 0.884, respectively, when LDCT-FBP images were used as reference. All of the calcified nodules were detected by both the observers on both the ULDCT-MBIR and ULDCT-FBP images. CONCLUSION: As compared with FBP, MBIR enables significant reduction of the image noise and artifacts and also better detection of noncalcified pulmonary nodules on ULDCT of the lung. Compared with LDCT-FBP images, ULDCT-MBIR images showed significantly reduced objective noise and comparable subjective image noise. Almost all of the noncalcified nodules and all of the calcified nodules could be detected on the ULDCT-MBIR images, when LDCT-FBP images were used as the reference.