Application of deep-learning reconstruction algorithm for enhanced CT scan of upper abdomen under different radiation doses: focus on noise, contrast-to-noise ratio and image quality

Objectives To compare the effect of deep-learning image reconstruction (DLIR) and adaptive statistical iterative reconstruction-Veo (ASiR-V) for upper abdominal-enhanced CT. The foci of this study were to validate whether the image effect of DLIR would be better than that of ASiR-V under the same dose and whether the image effect of DLIR under low dose would be not lower than that of ASiR-V under routine dose, and the two algorithms were studied from different layer thickness. Methods A total of 17 patients with hepatic lesions undergoing upper abdominal-enhanced CT examination were randomly divided into two groups: low dose (LD, n = 7) with 150 mAs and ultralow dose (ULD, n = 10) with 100 mAs; routine dose (RD) scans were subsequently performed in both groups ( n = 17). Images were reconstructed with DLIR with low, medium, and high strength (DL-L, -M, and -H, respectively) and ASiR-V with standard 30% and 70% (ASiR-V 30% and 70%) on thin (0.625 mm) and thick (5 mm) slices. Noise and contrast-to-noise ratio (CNR) measurements were performed. Two radiologists, blinded to examination details, scored four categories of image quality while comparing reconstructions for vascular significance, image noise and texture, lesion conspicuity and lesion diagnosis confidence. Results Under the same dose, DLIR had better performance in noise, CNR and image quality scores than ASiR-V did, DL-H performed the best. Under different radiation doses, ULD-DL-M did not differ significantly from RD-ASiR-V 30% in noise and CNR ( P > 0.05), but had lower image quality scores ( P < 0.05). Noise, CNR and image quality scores of the LD-DL-H were comparable to those of RD-ASiR-V 30% ( P > 0.05). Conclusion DLIR can significantly improve the image quality of enhanced abdominal CT under the same dose. By decreasing the radiation dose to 150mAs, DL-H showed comparable image quality to that of RD-ASiR-V 30%.