Comparison of Deep Learning Reconstructed HASTE Sequence and Conventional BLADE Sequence for Liver T2-weighted Imaging

Abstract Objectives To compare the acquisition time, image quality, and T2WI signal characteristics of the T2-weighted deep learning (DL) reconstructed HASTE sequence with conventional T2-weighted BLADE sequence for liver MRI. Methods From July 2023 to March 2024, 309 patients with suspected liver lesions were prospectively included. The Mann-Whitney U test and paired Wilcoxon signed-rank test were used to assess qualitative and quantitative parameters. Intra-class correlation coefficient (ICC) and kappa coefficients were used to assess agreement between the two readers. Results The median acquisition time of the T2-weighted DL HASTE sequence was shorter than that of the BLADE sequence (34 seconds [IQR, 34–34 seconds] vs 169 seconds [IQR, 158–181 seconds], respectively; P < 0.001). Artifacts (P < 0.001), pancreatic duct margin (P < 0.001), homogeneity of fat suppression (P = 0.002), strength of fat suppression (P < 0.001), and conspicuity of the liver smallest lesion (P < 0.001) were significantly better for the T2-weighted DL HASTE sequence. Contrast-to-noise ratio (CNR) of liver lesions was significantly higher for the DL HASTE sequence (P < 0.001), especially for cysts (P < 0.001). The two sequences showed similar performance in the T2WI signal characteristics of focal liver lesions. Conclusions The T2-weighted DL HASTE sequence led to shorter examination time and higher image quality with comparable T2WI signal characteristics of focal liver lesions compared with the BLADE sequence.