High-acceleration pancreatobiliary MRI with deep learning-based super-resolution reconstruction for evaluating presumed pancreatic intraductal papillary mucinous neoplasm

Sun Kyung Jeon ^1,2, Jeong Min Lee ^3,4,5, Junghoan Park ¹

⁰Department of Radiology, Seoul National University Hospital and Seoul National University College of Medicine, 101 Daehak‑ro, Jongno‑gu, Seoul, 03080, Korea.

Cancer Imaging : the Official Publication of the International Cancer Imaging Society +

|

October 25, 2025

View abstract on PubMed

Summary

Deep learning-based super-resolution (DL-SR) reconstruction significantly improved pancreatobiliary MRI quality for assessing pancreatic intraductal papillary mucinous neoplasms (IPMNs). This technique shows promise for accurately detecting malignant IPMN, enhancing diagnostic capabilities.

Area Of Science

Radiology
Medical Imaging
Artificial Intelligence in Medicine

Background

Pancreatic intraductal papillary mucinous neoplasms (IPMNs) require accurate assessment for malignancy.
Current pancreatobiliary MRI techniques face limitations in image quality and lesion conspicuity.

Purpose Of The Study

To evaluate the feasibility and diagnostic utility of a deep learning (DL)-based super-resolution (SR) reconstruction algorithm for pancreatobiliary MRI in assessing IPMNs.
To compare the image quality and diagnostic performance of DL-SR CS-VIBE with standard CS-VIBE for IPMN evaluation.

Main Methods

Retrospective analysis of 162 patients with presumed pancreatic IPMN (≥1 cm) who underwent pancreatobiliary MRI.
Acquisition of early and late portal venous phase (PVP) images using standard compressed sensing (CS)-VIBE and DL-SR CS-VIBE.
Assessment of image quality, lesion conspicuity, and diagnostic performance for malignant IPMN prediction using multi-reader multi-case analysis and ROC analysis.

Main Results

DL-SR CS-VIBE demonstrated significantly superior image quality and cystic lesion conspicuity compared to standard CS-VIBE (P < 0.001).
The area under the ROC curve (AUC) for predicting malignant IPMN using the full sequence including DL-SR CS-VIBE was 0.858.
Pooled sensitivity and specificity for malignant IPMN were 71.1% and 82.8%, respectively. Mural nodules ≥5 mm and main pancreatic duct size ≥10 mm showed high diagnostic accuracy (AUCs 0.736 and 0.720).

Conclusions

Pancreatobiliary MRI with DL-SR CS-VIBE enhances image quality and lesion conspicuity, improving the assessment of IPMNs.
The technique shows promising diagnostic accuracy for malignant IPMN.
Further research with larger cohorts is needed to validate findings and assess clinical impact.

Keywords:

Deep learning Intraductal papillary mucinous neoplasm MRI Pancreas Super-resolution