Gradient effect of KARL 3D iterative reconstruction on image noise in lung-window chest CT: a within-subject evaluation

Chest CT image quality is limited by image noise. Iterative reconstruction (IR) algorithms such as KARL 3D reduce noise while preserving detail, but the gradient effect of different KARL 3D levels on lung-window noise - where the sharp reconstruction kernel yields a high noise floor - and its dependence on patient body size remain incompletely characterized. In this retrospective within-subject study, 36 patients underwent chest CT (United Imaging uCT 780, 120 kV, mean CTDIvol 12.0 mGy). The same raw data were reconstructed with filtered back projection (FBP) and KARL 3D levels 3, 6, and 9. A small circular ROI (≈ 20 mm²) at the carina level (lung parenchyma posterior to the right main bronchus, avoiding vessels/bronchi) provided noise SD and SNR. Two radiologists scored four dimensions (noise, structure, artifact, acceptability) on a 5-point Likert scale; Reader A rated in two sessions for intra-reader reliability and Reader B rated once for inter-reader reliability. Repeated-measures ANOVA (rmANOVA; Greenhouse-Geisser) with Bonferroni-corrected paired t-tests assessed objective noise and SNR, and Friedman tests with Bonferroni-corrected Wilcoxon signed-rank tests assessed the subjective Likert scores; ICC and quadratic-weighted kappa assessed agreement; effective diameter (Deff) and subgroups (Mann-Whitney U) examined body-size effects. Noise decreased from 51.76 ± 13.31 HU (FBP) to 33.71 ± 8.84 HU (level 9; -34.9%), and SNR rose from 18.17 ± 6.58 to 27.61 ± 8.58 (all pairwise P < 0.001; rmANOVA SD F = 338.7, P < 0.001). Composite scores improved 3.21 ± 0.21 to 4.68 ± 0.15. Observer agreement was excellent for objective SD (ICC 0.970–0.983/0.970–0.975) and good-to-excellent for the subjective summed score (kappa 0.86–0.88 inter-reader; 0.79–0.96 intra-reader). Deff was not significantly correlated with noise (Pearson r = 0.07–0.14, all P > 0.05) and subgroups did not differ ( P = 0.70–0.75). KARL 3D produces a graded, level-dependent reduction in lung-window image noise; within the studied range of effective diameter, no significant association between the denoising effect and patient body size was observed. Higher reconstruction levels progressively reduced noise and improved subjective image quality, and the choice of level should be guided by the clinical task.