Improved upper abdominal MRI with VIBE sequences using deep learning-supported k-space sampling in a cohort undergoing MRI for gynecologic diseases.

Abstract

BackgroundVolumetric interpolated breath-hold examination (VIBE) is widely used in upper abdominal MRI but can be limited by low signal-to-noise ratio (SNR), especially when rapid acquisition is prioritized. Deep learning (DL)-enhanced reconstruction may improve image quality without extending acquisition time.PurposeTo compare subjective image quality, artifacts, noise, and estimated SNR and contrast-to-noise ratio (CNR) between DL-supported and standard (ST) VIBE sequences of the upper abdomen in women undergoing pelvic MRI, including the effect of contrast enhancement.Material and MethodsThis prospective study included 60 women (mean age 42.1 ± 14.7 years). Four axial sequence types were evaluated: ST and DL VIBE, both non-contrast (NC) and contrast-enhanced (CE). Three radiologists rated image quality, artifacts, and noise using a standardized 4-point Likert scale. Interobserver agreement and the effects of age and body mass index (BMI) were assessed. A quantitative region-of-interest (ROI)-based SNR/CNR analysis was also performed.ResultsDL VIBE yielded better image quality, fewer artifacts, and less noise than ST VIBE. DL-by-CE interactions were significant for image quality and artifacts. Interobserver agreement was moderate for image quality and artifacts but low for noise. Quantitative analysis showed no significant differences in estimated SNR or CNR between ST and DL VIBE before or after contrast administration (all p ≥ 0.35).ConclusionDL VIBE improves image quality and reduces artifacts, particularly in NC imaging.