Utility of super-resolution deep-learning reconstruction for improving visibility of hepatocellular carcinoma at hepatic dynamic CT: comparison with hybrid iterative reconstruction and conventional deep-learning reconstruction

Abstract Purpose The aim of this study was to evaluate whether Precise IQ Engine (PIQE), a super-resolution deep-learning reconstruction designed to enhance spatial resolution while reducing image noise using a deep convolutional neural network, improves the visibility of hepatocellular carcinoma (HCC) compared with hybrid iterative reconstruction (Adaptive Iterative Dose Reduction with three-dimensional processing [AIDR 3D]) and conventional deep-learning reconstruction (Advanced Intelligent Clear-IQ Engine [AiCE]) in hepatic dynamic CT. Methods and materials We evaluated 64 patients with HCC who had undergone hepatic dynamic CT. Hepatic arterial- and equilibrium phase (HAP and EP) scans were reconstructed with AIDR, AiCE, and PIQE. A radiologist measured the standard deviation of attenuation in the liver parenchyma as the image noise and calculated the contrast-to-noise ratio (CNR) = (ROI T - ROI L )/N in HAP and (ROI L - ROI T )/N in EP. ROI L is the mean attenuation of the liver parenchyma, ROI T the mean tumor attenuation, and N the noise. HCC visibility was assessed by 2 additional radiologists and graded with a focus on tumor margin delineation on a 5-point confidence scale. Results PIQE showed the lowest image noise, followed by AiCE and AIDR, in both HAP and EP (all p < 0.01). The CNR of HCCs was highest with PIQE, followed by AiCE and AIDR, in both HAP and EP (all p < 0.01). Qualitative visibility scores for HCCs were also highest with PIQE, followed by AiCE and AIDR, in both HAP and EP (all p < 0.01). Conclusion PIQE significantly improves HCC visibility compared with AIDR and AiCE in hepatic dynamic CT, along with reduced image noise and increased CNR.