Super-resolution deep learning-enhanced 2D MRCP improves main pancreatic duct visualization compared with 3D MRCP

The relative merits of 2D and 3D MRCP remain incompletely defined, and the effect of super-resolution deep learning reconstruction (SR-DLR) on main pancreatic duct (MPD) visualization remains unclear. We compared image quality and MPD visualization among 3D MRCP with SR-DLR (3D-SR), conventional 2D MRCP with compressed sensing (2D-CS), and 2D MRCP with SR-DLR (2D-SR), and exploratorily assessed diagnostic performance for MPD abnormalities.

This prospective study included 174 patients undergoing MRCP for pancreatic evaluation. Quantitative metrics included signal-to-noise ratio (SNR), contrast ratio (CR), and contrast-to-noise ratio (CNR) for the distal common bile duct and pancreatic cystic lesions. Two radiologists assessed image quality, MPD visualization, cystic lesion conspicuity, and MPD abnormalities.

3D-SR yielded the highest SNR, CR, and CNR for the distal common bile duct (all p ≤ 0.006). Overall image quality, motion artifact scores, MPD body/tail visualization, and margin sharpness were higher with 2D-SR and 2D-CS than 3D-SR, and higher with 2D-SR than 2D-CS (all p ≤ 0.037). Cystic lesion conspicuity was higher with 2D-SR and 2D-CS than 3D-SR, with the greatest difference for lesions ≤ 5 mm. All techniques correctly identified focal MPD narrowing with upstream dilatation (n = 3) and diffuse MPD dilatation (n = 13). For focal narrowing without upstream dilatation (n = 3), 3D-SR showed a higher false-positive rate (readers 1-2: 40.4-42.7%), whereas neither 2D technique showed false-positive findings.

In this surveillance cohort, 2D-SR provided the best overall MPD visualization; the lower false-positive rate for focal MPD narrowing with 2D techniques (only three true-positive cases) was exploratory and hypothesis-generating.