Automatic flow planning for fetal cardiovascular magnetic resonance imaging.

Background: Widening access to fetal flow imaging by automating real-time planning of two-dimensional (2D) phase-contrast flow imaging (OWL).

Methods: Two subsequent deep learning networks for fetal body localization and cardiac landmark detection on a coronal whole-uterus scan were trained on 167 and 71 fetal datasets, respectively, and implemented for real-time automatic planning of phase-contrast sequences. OWL was evaluated retrospectively in ten datasets and prospectively in seven fetal subjects (36+3-39+3 gestational weeks), with qualitative and quantitative comparisons to manual planning.

Results: OWL was successfully implemented in 6/7 prospective cases. Fetal body localization achieved a Dice score of 0.94±0.05, and cardiac landmark detection accuracies were 5.77±2.91 mm (descending aorta), 4.32±2.44 mm (spine), and 4.94±3.82 mm (umbilical vein). Planning quality was 2.73/4 (automatic) and 3.0/4 (manual). Indexed flow measurements differed by -1.8% (range -14.2% to 14.9%) between OWL and manual planning.

Conclusion: OWL achieved real-time automated planning of 2D phase-contrast cardiovascular magnetic resonance (CMR) for two major vessels, demonstrating feasibility at 0.55T with potential generalization across field strengths, extending access to this modality beyond specialized centers.