Application of AI-assisted compressed sensing in high resolution Gd-EOB-DTPA-enhanced MRI during hepatobiliary phase at 5.0 T

Liver tumor lesions are usually evaluated through magnetic resonance imaging (MRI), and currently the magnetic resonance field strengths are mainly 1.5 T and 3.0 T. There are relatively few reports on the application of 5.0 T ultra-high field MRI. This study aimed to quantitatively and qualitatively analyze the display and image quality of liver lesions using artificial intelligence (AI)-assisted compression sensing (ACS) combined with 5.0T ultra-high field MRI, and compare it with the traditional parallel imaging (PI) technique. This study also aimed to compare the image quality and lesion detection rate of hepatobiliary phase (HBP) imaging using PI and ACS of gadolinium ethoxybenzyl diethylenetriaminepentaacetic acid (Gd-EOB-DTPA)-enhanced MRI at 5.0 T ultra-high magnetic field. A total of 133 patients with suspected liver tumor lesions were recruited. All patients underwent 5.0 T ultra-high field Gd-EOB-DTPA-enhanced MRI, including axial three-dimensional (3D) fast spoiled gradient echo sequence using PI (PI-quick-3D), coronal PI-quick-3D, and axial quick-3D sequence using ACS during HBP. Two radiologists independently analyzed all images and recorded the location and number of lesions. To evaluate the image quality, objective scores (including signal-to-noise ratio, contrast-to-noise ratio, contrast ratio, and image sharpness) and subjective scores (including image artifacts, lesion margin clarity, liver edge clarity, vessel clarity, bile duct clarity, and overall image quality) were performed on both PI and ACS images. Weighted kappa test was performed to assess the consistency of subjective scores between the two readers. Either the chi-square or Wilcoxon signed-rank test was used to compare the differences in objective and subjective indices between the PI and ACS images. The scanning time between the PI and ACS methods was similar (17 vs. 18 s). Except for similar image artifacts scores, the other subjective and objective scores of axial and reconstructed coronal images of ACS-quick-3D were better than those of axial and coronal PI-quick-3D images (P<0.001). Compared with PI-quick-3D, the detection rate of lesions in ACS-quick-3D was similar (98.71% vs. 99.14% for lesions >10 mm, and 95.03% vs. 98.34% for lesions ≤10 mm, with P=0.614, and 0.279, respectively), whereas ACS had an advantage in detecting subcentimeter lesions located under the liver capsule. Compared to the traditional HBP imaging using PI, Gd-EOB-DTPA-enhanced MRI based on ACS-quick-3D could significantly improve the image quality with comparable lesion detection rate and without increasing the scanning time at 5.0 T.