Deep Learning-Accelerated Magnetic Resonance Enterography for the Assessment of Crohn's Disease

초록

Purpose: To compare image quality (IQ) between deep learning (DL)-accelerated magnetic resonance enterography (MRE) and conventional MRE in patients with Crohn's disease. Materials and Methods: From August 2023 to July 2024, we retrospectively reviewed two patient cohorts who underwent MRE using either conventional or DL-accelerated sequences. Two radiologists independently evaluated the overall IQ, sharpness, motion artifacts, and artificial appearance for single-shot fast spin echo (SSFSE), diffusion-weighted imaging (DWI), and precontrast and dynamic phases. Results: A total of 39 patients were included (median age, 31.0 years; conventional group, n = 15; DL group, n = 24). The mean overall IQ scores were 4.27 +/- 0.70 (conventional coronal SSFSE) and 4.79 +/- 0.41 (DL coronal SSFSE, p = 0.028), 3.00 +/- 0.65 (conventional DWI) and 3.92 +/- 0.65 (DL DWI, p = 0.001), and 3.60 +/- 1.06 (conventional arterial-phase T1-weighted imaging [T1WI]) and 4.33 +/- 0.70 (DL arterial-phase T1WI, p = 0.044). The DL group demonstrated higher overall IQ than the conventional group across all sequences. None of the examinations were classified as nondiagnostic (overall IQ <2) in either group. Across all phases, the DL group exhibited fewer motion artifacts but a more artificial appearance than the conventional group. Conclusion: DL-MRE achieved adequate IQ in patients with Crohn's disease while reducing scan time.

키워드