Deep learning reconstruction for T2-weighted and contrast-enhanced T1-weighted magnetic resonance enterography imaging in patients with Crohn's disease: Assessment of image quality and clinical utility

Abstract

Purpose: To investigate the image quality of deep learning-reconstructed T2-weighted half-Fourier single-shot turbo spin echo (DL T2 HASTE) and contrast-enhanced T1-weighted volumetric interpolated breath-hold examination (DL T1 VIBE) of magnetic resonance enterography in patients with Crohn's disease. Methods: Ninety-three patients for the comparison between T2 HASTE and DL T2 HASTE, and forty-two patients for the comparison between T1 VIBE and DL T1 VIBE were enrolled. Two reviewers assessed image quality using 5-point scale scoring, and subjective reviewer preferences. Acquisition times were also compared. The diagnostic performance for the detection of active inflammation was evaluated in three bowel segments (ileum, terminal ileum, and right colon). Results: The interobserver agreement on image quality was better for T2 HASTE (0.624, standard error 0.083), and T1 VIBE (0.53, SE 0.113) than for DL T2 HASTE (0.443, SE 0.111) and DL T1 VIBE (0.173, SE 0.153). The mean image quality scores showed no significant differences between T2 HASTE and DL T2 HASTE images, as well as between T1 VIBE and DL T1 VIBE images (p > 0.05). Both reviewers preferred T2 HASTE over DL T2 HASTE, although the preference between T1 VIBE and DL T1 VIBE was controversial. The acquisition times were shorter for DL T2 HASTE (16.17 +/- 2.53 s) and DL T1 VIBE (14.19 +/- 0.84 s) compared to T2 HASTE and T1 VIBE. The per-segment diagnostic performance for detecting active inflammation was identical between T2 HASTE and DL T2 HASTE. Conclusion: DL-reconstruction allows one-breath hold scan while maintaining image quality in both T2 HASTE and T1 VIBE images.