Abstract
Objective
To evaluate the performance of the Deep Learning reconstruction algorithm Precise Image (PI) in improving abdominal computed tomography (CT) image quality and its diagnostic efficacy for adrenal and renal lesions.
Methods
A total of 191 patients who underwent abdominal CT scan were retrospectively enrolled. All image datasets were reconstructed into four sequences with a slice thickness of 1 mm: Filtered back projection (FBP), Iterative Reconstruction (iDose), PI Standard, and PI Smooth. We measured CT attenuation values and the standard deviation (SD) of adrenal glands, renal parenchyma, and intra‐abdominal fat for each reconstruction group. The contrast‐to‐noise ratio (CNR) and signal‐to‐noise ratio (SNR) of adrenal and renal parenchyma were subsequently calculated. Two radiologists independently performed subjective assessments for overall image quality and diagnostic confidence for adrenal nodules and renal cysts using the Likert 5 and 4 scales, respectively. Inter‐reader agreement was quantified via the Kappa test. Intergroup comparisons were conducted using the Kruskal–Wallis test with Bonferroni post hoc correction. A priori power analysis was performed to verify the adequacy of the sample size.
Results
Statistically significant intergroup differences were identified for all subjective evaluation metrics (
p
p
p
0.99 for all primary clinical endpoints, confirming robust and reproducible statistical findings.
Conclusion
Compared with FBP and iDose, the PI reconstruction algorithm substantially improves abdominal CT image quality. PI Smooth delivers the highest diagnostic confidence for small lesions and renal cysts, providing robust quantitative and qualitative support for precise clinical evaluation.