Image quality, diagnostic performance of reduced-dose abdominal CT with artificial intelligence model-based iterative reconstruction for colorectal liver metastasis: a prospective cohort study.

Background: The optimization of regularization strategies in computed tomography (CT) iterative reconstruction may allow for a reduced dose (RD) without compromising image quality, thus the diagnostic ability of RD imaging must be considered, especially for low-contrast lesions. In this study, we evaluated the image quality and diagnostic performance of 50% RD CT for low-contrast colorectal liver metastasis (CRLM) with artificial intelligence model-based iterative reconstruction (AIIR) and standard-dose (SD) CT with hybrid iterative reconstruction (HIR). Methods: In this prospective study, consecutive participants with pathologically proven colorectal cancer and suspected liver metastases who underwent portal venous phase CT scans both at SD and RD between June and November 2022 were included. All images were reconstructed by HIR and AIIR. Two radiologists detected and characterized liver lesions with RD HIR, SD HIR, and RD AIIR and scored the image quality. The contrast-to-noise ratio (CNR) for metastases were recorded. The diagnostic performance for CRLM of each reconstruction algorithm was analyzed and compared using the receiver operating characteristic curve and the area under the curves (AUC). Results: A total of 56 participants with 422 liver lesions were recruited. The mean volume CT dose indices of the SD and RD scans were 9.5 and 4.8 mGy. RD AIIR exhibited superior subjective image quality and higher CNR for liver metastases than did RD/SD HIR. In all liver lesions and lesions ≤10 mm, the detection rates of RD AIIR (83.3% and 71.5%) were both significantly higher than those of RD HIR (76.3% and 62.4%; P=0.002 and P=0.003); meanwhile, they were similar to those of SD HIR (81.4% and 69.6%; P=0.307 and P=0.515). The AUCs of RD AIIR for all liver lesions and lesions ≤10 mm (0.858 and 0.764) were greater than those of RD HIR (0.781 and 0.661; P<0.001) and were similar to those of SD HIR (0.863 and 0.762; P=0.616 and 0.845). Conclusions: AIIR can improve CT image quality at 50% RD while preserving diagnostic performance and confidence for low-contrast CRLM in all lesions and lesions ≤10 mm and may thus serve as a promising tool for follow-up monitoring in patients with colorectal cancer while inflicting less radiation damage.