Exploring the diagnostic performance of low-dose CT for colorectal cancer assessment: feasibility of artificial intelligence iterative reconstruction

Background: Contrast-enhanced long-range abdominal computed tomography (CT) plays a crucial role in the diagnosis and staging of colorectal cancer (CRC) yet may involve excessive radiation exposure, especially for patients requiring multiple examinations. This raises the need to reduce radiation dose while maintaining image quality. Given the potential of a novel artificial intelligence iterative reconstruction (AIIR) algorithm to improve image quality in low-dose (LD) CT, this study aimed to investigate the feasibility of applying AIIR to LD abdominal CT for CRC diagnosis.

Methods: In this prospective study, 203 patients with pathology-confirmed CRC underwent abdominal CT, including a routine-dose (RD) scan (120 kVp, 200 mAs) followed by a LD scan (120 kVp, 20 mAs) at the portal venous phase (PVP). RD images were reconstructed with hybrid iterative reconstruction (HIR, RD-HIR) and AIIR (RD-AIIR), whereas LD images were reconstructed with AIIR (LD-AIIR). Diagnostic performance for assessing visceral peritoneal invasion (VPI) and regional lymph node metastasis (RLNM) was characterized using receiver operating characteristic (ROC) analysis. Qualitative image quality was rated using a five-point scale, and tumor contrast-to-noise ratio (CNR) was measured.

Results: The mean effective dose (ED) of LD scan was 90.3% lower than that of RD scan (1.5±0.2 vs. 14.9±2.4 mSv). RD-AIIR achieved significantly higher area under the curve (AUC) and accuracy for diagnosing VPI (0.89 and 91.13%, respectively) and RLNM (0.72 and 71.92%, respectively) compared to the other reconstructions (all P<0.05). LD-AIIR showed comparable AUC and accuracy to RD-HIR for VPI (0.81 vs. 0.80 and 82.96% vs. 78.33%, respectively; both P>0.05), but demonstrated inferior performance for RLNM (0.65 vs. 0.68 and 66.51% vs. 68.47%, respectively; P<0.05). AIIR significantly improved tumor CNR (P<0.05) and the qualitative image quality was comparable between LD-AIIR and RD-HIR (P>0.05).

Conclusions: AIIR offers superior image quality compared to HIR. AIIR allows up to 90.3% dose reduction for reliable VPI assessment in CRC while maintaining comparable image quality to that of RD-HIR on abdominal CT.