Abstract
Background: While deep learning image reconstruction(DLIR) has been applied successfully in thoracic, abdominal, and vascular examinations, its application in low-dose renal CT protocols has not been previously explored.
Purpose: To explore the value of DLIR in reducing radiation dose and improving image quality in contrast-enhanced renal CT compared with the adaptive statistical iterative reconstruction Veo(ASIR-V).
Material and Methods: Methods: 129 renal disease patients underwent unenhanced and triphasic-enhanced CT scans, utilizing a standard 120 kVp dose for parenchymal-phase scans and a lower 100 kVp dose for corticomedullary-phase scans. Images in both phases were reconstructed with high-strength DLIR(DLIR-H), medium-strength DLIR(DLIR-M) and ASIR-V level 50%(ASIR-V-50%) for comparison. CT values and standard deviations were measured and compared for various tissues in both phases, and two radiologists assessed image quality using a 5-point Likert scale in seven aspects.
Results: A total of 118 patients were included, with corticomedullary-phase radiation dose reduced by over 15% compared to parenchymal-phase (CTDIvol: 6.57±2.13mGy vs. 7.75±2.63mGy). DLIR-M and DLIR-H exhibited significantly lower image noise in both phases compared to ASIR-V-50% (p<0.001). Corticomedullary-phase DLIR-M and DLIR-H images reduced subcutaneous-adipose tissue noise by 15% and 40% compared to parenchymal-phase ASIR-V-50%. Subjectively, DLIR-H (4.16±0.62) and DLIR-M (3.76±0.68) using 100 kVp outperformed ASIR-V-50% (3.42±0.52) at 120 kVp (p<0.001).
Conclusion: DLIR-H and DLIR-M significantly reduce image noise and generate images with better image quality and diagnostic confidence with a 15% dose reduction than ASIR-V-50%.
Clinical Trial Number
2023-278, First Hospital of Jilin University, Changchun, China.