Image Quality Improvement of Low-dose Abdominal CT Using Deep Learning Image Reconstruction Compared With the Second Generation Iterative Reconstruction.

Abstract

Whether deep learning-based CT reconstruction could improve lesion conspicuity on abdominal CT when the radiation dose is reduced is controversial. To determine whether DLIR can provide better image quality and reduce radiation dose in contrast-enhanced abdominal CT compared with the second generation of adaptive statistical iterative reconstruction [ASiR-V]. This study aims to determine whether deep-learning image reconstruction [DLIR] can improve image quality. In this retrospective study, a total of 102 patients were included, who underwent abdominal CT using a DLIR-equipped 256-row scanner and routine CT of the same protocol on the same vendor's 64-row scanner within four months. The CT data from the 256-row scanner were reconstructed into ASiR-V with three blending levels [AV30, AV60, and AV100], and DLIR images with three strength levels [DLIR-L, DLIR-M, and DLIR-H]. The routine CT data were reconstructed into AV30, AV60, and AV100. The contrast-to-noise ratio [CNR] of the liver, overall image quality, subjective noise, lesion conspicuity, and plasticity in the portal venous phase [PVP] of ASiR-V from both scanners and DLIR were compared. The mean effective radiation dose of PVP of the 256-row scanner was significantly lower than that of the routine CT [6.3±2.0 mSv vs. 2.4±0.6 mSv; p< 0.001]. The mean CNR, image quality, subjective noise, and lesion conspicuity of ASiR-V images of the 256-row scanner were significantly lower than those of ASiR-V images at the same blending factor of routine CT, but significantly improved with DLIR algorithms. DLIR-H showed higher CNR, better image quality, and subjective noise than AV30 from routine CT, whereas plasticity was significantly better for AV30. DLIR can be used for improving image quality and reducing radiation dose in abdominal CT, compared with ASIR-V.