Improved image quality and dose reduction in liver CT using deep learning-based reconstruction: A comparative study

To evaluate and compare the image quality and radiation dose of CT images reconstructed with Deep Learning-based Image Reconstruction (DLIR) and Hybrid Iterative Reconstruction (HIR) algorithms in patients with focal liver lesion(s).

A total of 153 patients with at least one hepatic lesion (including hepatic cysts, hemangiomas, hepatocellular carcinomas, and cholangiocarcinomas) underwent two CT scans. The scans were performed using two different scanners, which employ DLIR and HIR, respectively. Image quality was evaluated using a 5-point Likert scale. CT attenuation values, expressed in Hounsfield units (HU), were recorded within manually drawn regions of interest (ROI) in liver lesions, liver parenchyma, paraspinal muscle (latissimus dorsi), aortic lumen, portal vein lumen, and adipose tissue, measured at the level of the subcutaneous fat of the anterior abdominal wall, at the axial plane corresponding to the bifurcation of the portal vein. Image noise, expressed as ROI's standard deviation (SD), signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR) were calculated. Radiation dose was assessed through the Computed Tomography Dose Index (CTDI) and Dose Length Product (DLP) values.

A total of 306 hepatic lesions were analyzed. The overall median image quality score for CT images reconstructed using the DLIR algorithm was significantly higher than that for HIR [4 (range 4-5) vs. 4 (range 3-4), p < 0.001]. CT attenuation values were significantly higher, and SD values significantly lower, in all anatomic regions in images reconstructed with DLIR compared to HIR (all p < 0.001). SNR and CNR values were higher in CT images reconstructed with DLIR, with significant differences observed across all anatomical regions (all p < 0.001). Additionally, the radiation dose was significantly lower in the DLIR compared to the HIR group (p < 0.001).

DLIR achieves a significant reduction in radiation dose for liver CT imaging, while also improving image quality both qualitatively and quantitatively.