Comparison of 2D and 3D fast non-local means in pediatric abdominal low-dose computed tomographic image

Fast non-local means (FNLM) is a post-processing technique that suppresses noise while preserving edges. FNLM can be implemented in two-dimensional (2D) and three-dimensional (3D) forms, yet the effect of smoothing factor selection on performance in pediatric low-dose computed tomography (CT) remains unclear. This study applied 2D and 3D FNLM to 30 pediatric abdominal CT cases with simulated low-dose noise using smoothing factors from 0.001 to 0.060. Image quality was evaluated using the coefficient of variation (COV), peak signal-to-noise ratio (PSNR), and structural similarity index measure (SSIM). In both algorithms, COV decreased markedly as the smoothing factor increased, particularly from 0.001 to 0.016, whereas PSNR and SSIM showed relatively limited variation. SSIM reached the maximum value at 0.016, followed by a gradual decrease. Linear mixed-effects models demonstrated significant effects of the smoothing factor on all metrics. Wilcoxon signed-rank test indicated significantly lower COV and higher SSIM with 3D FNLM while PSNR showed no significant difference. 3D FNLM provided significant improvements in noise uniformity and structural preservation with modest effects. In contrast, 2D FNLM remains a practical option when computational efficiency takes priority. Overall, a smoothing factor around 0.016 offers a favorable balance between noise reduction and structural fidelity.