Comparación de la calidad de imagen entre la reconstrucción de aprendizaje profundo de cortes finos a dosis bajas y...

To compare image quality between low-dose thin-slice deep-learning reconstruction (DLR) and standard-dose thick-slice hybrid iterative reconstruction (HIR) in pediatric abdominal CT.

82 children (≤6 years) who underwent contrast-enhanced abdominal CT with standard-dose (STD, n = 41) or low-dose (LD, n = 41) protocol matched by age and weight were retrospectively identified. STD and LD images were reconstructed at 3.0-mm using HIR (STD-HIR/3.0 and LD-HIR/3.0, respectively). LD images were also reconstructed at 0.5-mm using HIR (LD-HIR/0.5) and DLR (LD-DLR/0.5). Size-specific dose estimate (SSDE) was compared between groups. Image noise and contrast-to-noise ratio (CNR) were quantified. Noise power spectrum (NPS) and edge-rise slope (ERS) were employed for noise texture and edge sharpness measures, respectively. For subjective evaluation, noise magnitude, noise texture, edge sharpness, delineation of small structures, and diagnostic confidence were rated on a 5-point scale (1 = undiagnostic, 5 = best).

SSDE was on average 59.5% lower in LD than in STD group (1.7 ± 0.4 vs. 4.2 ± 0.8 mGy, p < 0.001). Image noise was lower in LD-DLR/0.5 compared to STD-HIR/3.0, LD-HIR/3.0, and LD-HIR/0.5 (8.7 ± 1.5, 9.6 ± 1.1, 11.7 ± 1.9, and 17.9 ± 2.3 HU, respectively, all p ≤ 0.017). LD-DLR/0.5 showed equivalent CNR (e.g., liver CNR: 7.1 ± 2.1 vs. 7.0 ± 2.8, p = 0.827) and higher ERS (64.3 ± 23.9 vs. 53.3 ± 10.6 HU/mm, p = 0.011) with similar average NPS frequency (0.281 ± 0.042 vs. 0.291 ± 0.027 mm^-1, p = 0.177) compared to STD-HIR/3.0. Subjective scores for all criteria were higher in LD-DLR/0.5 than in STD-HIR/3.0 (e.g., diagnostic confidence score: 4.5 ± 0.5 vs. 3.4 ± 0.4, p < 0.001).

Low-dose thin-slice DLR improved edge sharpness, small structure visualization, and diagnostic confidence in pediatric abdominal CT without increasing noise compared to standard-dose thick-slice HIR.