Volumetric Auto-Segmentation of the Pancreaticobiliary System for Evaluating MRCP Image Quality: Efficacy Before and After Contrast-Enhanced

Motivation: An automated segmentation model of the pancreaticobiliary system facilitates the acquisition and assessment of diagnostically valuable MRCP images. Goal(s): To determine impact of contrast enhancement and varying scanning protocols on image quality and develop automated segmentation models of the pancreaticobiliary ducts. Approach: A multi-center retrospective study trained and validated nn-UNet models, comparing MRCP protocol quality pre- and post-contrast. Results: The nn-UNet models trained on pre-contrast, post-contrast, and combined MRCP images achieved consistent segmentation accuracy across datasets (DICE>0.80). Post-contrast-RTr-MRCP showed improved contrast and reduced SNR (P<0.001), along with a higher percentage of 3-10 mm duct diameters (P<0.05), while no significant volume changes were observed. Impact: The automatic segmentation model provides a robust tool for efficient 3D pancreaticobiliary reconstruction, improving MRCP workflow. Future studies may investigate optimizing MRCP quality based on patient factors or clinical scenarios, and assess diagnostic value of quantitative parameters extracted from segmentation.