Comparative Analysis of Bilateral-Based Noise Reduction Algorithms in Low-Dose Pediatric Computed Tomographic Images

Background: Noise reduction in low-dose pediatric computed tomography (CT) requires effective suppression of quantum noise without compromising subtle anatomical structures. This study compared three bilateral-based noise reduction algorithms, namely the conventional bilateral filter (BF), adaptive BF (ABF), and trainable BF (TBF), using eight low-dose pediatric abdominal CT cases. Methods: Denoising performance was evaluated in the axial and coronal planes through visual assessment and quantitative measurements of the coefficient of variation (CV) within the liver parenchyma and the contrast-to-noise ratio (CNR) between the liver and portal vein. Statistical comparisons were performed using the Friedman test, followed by the Wilcoxon signed-rank tests with Holm-adjustment. All three methods reduced the visible noise present in the original images. Results: ABF improved uniformity in homogeneous liver regions while preserving anatomical detail, whereas TBF produced the strongest apparent noise suppression but introduced greater blurring and occasional ambiguous structures. Quantitatively, the CV decreased progressively from BF to ABF and TBF. The TBF method achieved the lowest average CV and the highest average CNR. Significant differences were observed among the four conditions for both metrics in each imaging plane. However, CNR differences between the three denoising methods were less consistent, particularly for comparisons involving TBF. Conclusions: Overall, ABF provided a favorable balance between noise reduction and structural preservation, whereas TBF offered stronger image specific denoising with additional concerns regarding signal fidelity and reproducibility. Further refinement is required to ensure clinical reliability without compromising quantitative accuracy or anatomical details.