Utility of breath-hold half-Fourier single-shot turbo spin-echo with deep learning-based reconstruction for acquiring fat-suppressed T2-weighted images of the pancreas.

Purpose: To determine the best technique for acquiring fat-suppressed T2-weighted images (FS-T2WI) of the pancreas, in terms of acquisition time and image quality, among conventional respiratory-gated turbo spin echo (RG-TSE), research breath-hold TSE with deep learning-based reconstruction (BH-DL-TSE), and research breath-hold half-Fourier single-shot turbo spin-echo with deep learning-based reconstruction (BH-DL-HASTE). Methods: Eighty-seven patients with suspected pancreatic diseases who underwent magnetic resonance imaging between April and May 2022 were evaluated. Sixty-four focal pancreatic lesions > 5 mm were also evaluated. FS-T2WI of the pancreas were acquired with 3-T scanners using RG-TSE, BH-DL-TSE, and BH-DL-HASTE. Two radiologists qualitatively evaluated image quality on a 5-point scale. We calculated the pancreas signal-to-noise ratio (SNR) and lesion-to-pancreas contrast-to-noise ratio (CNR). Friedman's test and Dunn's multiple comparisons test were performed to assess differences among the three techniques. Results: The mean acquisition time was 3 min 51 s (95 % confidence interval: 3 min and 37 s - 4 min and 6 s) for RG-TSE, 40 s for BH-DL-TSE, and 20 s for BH-DL-HASTE. BH-DL-HASTE provided the fewest artifacts and the sharpest edges (P < 0.0001), the most conspicuous main pancreatic duct (P = 0.0001 versus RG-TSE and P < 0.0001 versus BH-DL-TSE), and more conspicuous lesions than those in BH-DL-TSE (P = 0.0028). There was no significant difference in the SNR (P = 0.0766). RG-TSE and BH-DL-HASTE provided a higher CNR than BH-DL-TSE (P = 0.0018 and P = 0.0026, respectively). Conclusions: BH-DL-HASTE would be better than RG-TSE and BH-DL-TSE for acquiring FS-T2WI of the pancreas.