3-dimensional gradient and spin-echo magnetic resonance cholangiopancreatography with deep learning reconstruction at 3T: Achieving superior image quality with reduced acquisition time.
Authors
Affiliations (3)
Affiliations (3)
- Department of Radiology, Hamamatsu University School of Medicine, 1-20-1, Handayama, Chuo-ku, Hamamatsu City, Shizuoka 431-3192, Japan. Electronic address: [email protected].
- Department of Radiology, Hamamatsu University School of Medicine, 1-20-1, Handayama, Chuo-ku, Hamamatsu City, Shizuoka 431-3192, Japan.
- Philips Japan, Azabudai Hills Mori JP Tower 15F, 1-3-1 Azabudai, Minato-ku, Tokyo 106-0041, Japan.
Abstract
To evaluate the image quality and clinical feasibility of breath-hold 3D-magnetic resonance cholangiopancreatography (MRCP) using a gradient and spin-echo (GRASE) technique with deep learning reconstruction (GRASE-DLR) versus GRASE without DLR and MRCP using a turbo spin-echo sequence with DLR (TSE-DLR) at 3 T. Sixty-five consecutive patients who underwent MRCP were enrolled retrospectively. Three MRCP protocols were compared: GRASE-DLR, GRASE without DLR, and TSE-DLR. The acquisition time, image quality metrics, and diagnostic performance were assessed. Two radiologists independently performed qualitative and diagnostic assessment. Subgroup analysis was performed for patients with poor breath-hold capacity (n = 8). GRASE-DLR reduced the acquisition time by 45.4% and 49.1% compared with TSE-DLR and GRASE, respectively (p < 0.001). GRASE-DLR yielded high overall image quality (p < 0.001) and artefact reduction (p < 0.001 vs. TSE-DLR; p < 0.05 vs. GRASE), and improved major duct visualization (p < 0.001). Diagnostically, GRASE-DLR achieved higher sensitivity for biliary disease (93.8% vs. 56.2-62.5% for TSE-DLR) and showed a trend toward higher sensitivity for pancreatic disease (87.5% vs. 55.0-72.5%). Accuracy ranged from 89.5 to 94.7% (biliary) and 84.8-87.0% (pancreatic). The area under the receiver operating characteristic curve was higher with GRASE-DLR for biliary (0.802-0.969) and pancreatic disease (0.854-0.890). Indeterminate biliary anatomical variation findings were nearly eliminated with GRASE-DLR (0-1.5% vs. 21.5-26.2% for TSE-DLR). In patients with poor breath-hold capacity, GRASE-DLR demonstrated pronounced improvements in image quality and artefact reduction. GRASE-DLR provides favorable image quality with substantially reduced acquisition time and improved diagnostic performance, particularly in patients with poor breath-hold tolerance.