Effect of iterative reconstruction on image quality of low-dose chest computed tomography

AIM: To assess quality and radiologists' preference of low-dose computed tomography (LDCT) reconstructed with filtered back projection (FBP) or Iterative Reconstruction. METHODS: Thin-section LDCTs (1-mm thick contiguous images; 120 kVp; 30 mAs) of 38 consecutive unselected patients, evaluated for various clinical indications, were reconstructed by four different reconstruction algorithms: FBP and Sinogram-AFfirmed Iterative Reconstruction (SAFIRE) with three different strengths, from 2 to 4 (i.e. S2, S3, S4). The image noise was recorded. Two thoracic radiologists visually compared both anatomic structures (interlobular septa, lung fissures, centrilobular artery, bronchial wall, and small vessels) and lung abnormalities (intralobular reticular opacities, nodules, emphysema, cystic lung disease, decreased-attenuation areas related to constrictive obliterans bronchiolitis, patchy ground-glass opacity, consolidation, and bronchiectasis) using a qualitative four-point scale grading system of the image quality. RESULTS: A lower amount of noise was recorded for LDCTs reformatted with any SAFIRE algorithm, as compared to FBP (P < 0.0001). The noise levels decreased as the SAFIRE strength increased from S2 to S4. The visual score of the subsegmental/segmental bronchial wall was greater for the FBP datasets compared to any SAFIRE dataset (P < 0.0001 for reviewer 1; P < 0.02 for reviewer 2). The decreased lung attenuation pattern score was lower on the S4 images for one reviewer, as compared to the other LDCT datasets (P = 0.003). No other differences in terms of radiologists' preference were recorded among FBP, S2, S3, and S4. Interobserver agreement was moderate only for fissures and bronchial wall, and good to excellent for the remainders. CONCLUSION: Iterative reconstructions showed lower image noise but did not provide any real improvement for the radiologists' evaluation of thin-section LDCT of the lung.