Comparison of Knowledge-based Iterative Model Reconstruction (IMR) with Hybrid Iterative Reconstruction (iDose4) Techniques for Evaluation of Hepatocellular Carcinomas Using Computed Tomography.

RATIONALE AND OBJECTIVES To compare tumor conspicuity of small hepatocellular carcinomas (HCCs) and image quality on knowledge-based iterative model reconstruction low-dose computed tomography (IMR-LDCT) with hybrid iterative reconstruction standard-dose CT (iDose4-SDCT). METHODS Thirty-two patients (mean age 61.9 ± 9.7 years; male:female 27:5; mean body mass index 25.6 ± 3.8 kg/m2) with cirrhosis and 40 HCCs in IMR-LDCT group and 33 patients (mean age 60.1 ± 7.4 years; male:female 28:5; body mass index 26.7 ± 3.2 kg/m2) with cirrhosis and 40 HCCs in iDose4-SDCT group were included in this retrospective study. Objective analysis of reconstructed iDose4 and IMR images was done for contrast-to-noise ratio of HCCs (CNRHCC), image noise, signal-to-noise ratio of portal vein (SNRPV), and inferior vena cava (SNRIVC). Subjective analysis of tumor conspicuity and image quality was done by two independent reviewers in a blinded manner. Mean volume CT dose index, dose length product, and effective dose for both groups were compared. RESULTS The CNRHCC was significantly higher in IMR-LDCT compared to iDose4-SDCT in both arterial phase (AP), p < 0.0001, and delayed phase (DP), p < 0.0001. Image noise was significantly lower in IMR-LDCT compared to iDose4-SDCT in AP, portal venous phase, and DP with p < 0.0001. IMR-LDCT showed significantly higher SNRPV (p < 0.0001) and SNRIVC (p < 0.0001) compared to iDose4-SDCT. On subjective analysis, IMR-LDCT images showed better image quality in AP, portal venous phase, and DP and better tumor conspicuity in AP and DP. IMR-LDCT (21.4 ± 4.6 mSv) achieved 36.9% reduction in the effective dose compared to iDose4-SDCT (33.9 ± 6.2 mSv). CONCLUSION IMR algorithm provides better image quality and tumor conspicuity with considerable decrease in image noise compared to iDose4 reconstruction technique even on LDCT.