Uncertainty-aware Coarse-to-Fine Main Pancreatic Duct Segmentation for Width Measurement in 3D MRCP

Zipei Zhang; Jaakko Sahlsten; Mikko Saukkoriipi; Johanna Ronkainen; Mona Lehto; Akseli Aittoniemi; Irina Rinta-Kiikka; Kimmo Kaski

cs.CVOct 29, 2026

Uncertainty-aware Coarse-to-Fine Main Pancreatic Duct Segmentation for Width Measurement in 3D MRCP

Authors: Zipei Zhang, Jaakko Sahlsten, Mikko Saukkoriipi, Johanna Ronkainen, Mona Lehto, Akseli Aittoniemi, Irina Rinta-Kiikka, Kimmo Kaski

Organizations: Department of Computer Science, Aalto University School of Science, Finland · Department of Radiology, Tampere University Hospital, Wellbeing Services County of Pirkanmaa, Finland · Faculty of Medicine and Health Technology, Tampere University, Finland

Abstract

Assessment of the main pancreatic duct (MPD) from volumetric magnetic resonance cholangiopancreatography (MRCP) scans is an essential task for ductal dilatation diagnosis and pancreatic cancer analysis. However, manual assessment is affected by reader variability in which automated analysis may assist. Nevertheless, fully automated and uncertainty-aware MPD analysis from volumetric MRCP remains underexplored area of research. In this study, we propose an automatic and uncertainty-aware deep learning-based pipeline for MPD segmentation and maximum-width analysis from MRCP scans. A coarse localisation model first identifies the ductal region-of-interest, followed by a fine model that segments the MPD within the region. The dataset consists of 256 MPD-segmented MRCP scans from 69 patients. A randomly selected holdout test subset of 14 patients with 55 MRCP volumes was used for evaluation. We employed a five-model cross-validation ensemble for improved performance and uncertainty estimation. The best MPD segmentation performance was 0.634 DSC, 0.744 precision, and 0.593 recall. The mean absolute error for maximum MPD width prediction was 0.355~mm and dilated MPD ( $\geq 5$ ~mm) was detected with 75.0% sensitivity and 74.4% specificity. These novel results suggest that automated MPD segmentation can be used to aid MRCP-based duct measurements.

Figures & tables

Characteristics	Development Dataset	Test Dataset
Number of MRCP volumes	201	55
Patient demographics
Number of individuals	55	14
Age, mean (SD), years	72.4 (9.3)	75.6 (7.9)
Male : Female	44 (80.0%) : 11 (20.0%)	5 (35.7%) : 9 (64.3%)
Normal MPD	44 (80.0%)	10 (71.4%)

Table 1: Characteristics of Development and Test Datasets. In our dataset, main pancreatic duct (MPD) dilation is caused by MPD Intraductal Papillary Mucinous Neoplasm (IPMN) or mix type IPMN.

Method	DSC $\uparrow$	Precision $\uparrow$	Recall $\uparrow$	ASSD (mm) $\downarrow$	HD95 (mm) $\downarrow$	Zero prediction mask $\downarrow$
DynUNet	0.579	0.613	0.583	3.50	23.7	0 ( 0.00% )
3D U-Net + 3D U-Net	0.629	0.717	0.585	4.65	32.0	0 ( 0.00%)
3D U-Net + Swin UNETR	0.516	0.634	0.479	11.7	45.5	1 (1.82%)
3D U-Net + 3D SegResNet	0.634	0.744	0.593	4.18	22.9	1 (1.82%)

Table 2: Main pancreatic duct segmentation performance on the test set volumes (N=55). Best results are shown in bold.

Figure 1: Model performance across uncertainty threshold on referral based approach. The most uncertain cases are referred and average performance on remaining cases is plotted. In subfigures A. to C. the pipelines are compared using the best-performing uncertainty measure coefficient of variation (CV), and in subfigures D. to F. the three uncertainty measures predictive entropy (H), structure-based predictive entropy (SH), and coefficient of variation (CV), are compared using the best performing model i.e., 3D U-Net + 3D SegResNet. Note that evaluation for two of the pipelines start at 98% data remaining on the distance based measures due to zero prediction mask.

Figure 2: Correlation analysis between uncertainty measure and model performance. Subfigures A. to C. are scatterplots with fitted linear line between quantitative measures and coefficient of variation (CV) and subfigures D. to F. are boxplots with ordinal qualitative measures and CV.

Figure 3: Detailed look at 3D U-Net + 3D SegResNet performance within each test-set patient ordered ascending by median DSC performance. Segmentation performance is shown on subfigures A. to C. while subfigures D. and E. show qualitative measures by radiologists. Subfigure F. shows the coefficient of variation uncertainty measure.

Medical Image Segmentation

Medical image segmentation delineates organs, lesions, and other structures in scans to support measurement, diagnosis, treatment planning, and research.

See all 796 papers

CrossPan: A Comprehensive Benchmark for Cross-Sequence Pancreas MRI Segmentation and Generalization
Apr 20, 2026Linkai Peng, Cuiling Sun, Zheyuan Zhang +10Image Segmentation Magnetic Resonance Imaging
Department of Radiology, Northwestern University, Chicago, USA. · Department of Electrical and Computer Engineering, Northwestern University, Evanston, USA. · Department of Computer Science, Northwestern University, Evanston, USA. +4
Multimodal Assessment of Pancreatic Cancer Resectability Using Deep Learning
Jul 15, 2026Vincent Ochs, Christoph Kuemmerli, Florentin Bieder +123D Medical Imaging Multimodal Classification
University of Basel, Department of Biomedical Engineering, 4123 Allschwil, Switzerland · Clarunis, University Digestive Health Centre, Basel, Switzerland · Department of General Surgery, Kantonsspital Aarau, Tellstrasse, 5000, Aarau, Switzerland +3
Assessing Pancreatic Ductal Adenocarcinoma Vascular Invasion: the PDACVI Benchmark
Apr 30, 2026M. Riera-Marín, O. K. Sikha, J. Rodríguez-Comas +23Inter-Rater Reliability Medical Image Analysis
BCN Medtech, Universitat Pompeu Fabra, Barcelona, Spain · Universitätsklinikum Erlangen, Department of Radiology of the Uniklinikum Erlangen (UKER), Erlangen, Germany · University Hospital Erlangen, Imaging Science Institute, Erlangen, Germany +10

Apr 20, 2026J/K move · Enter open · S save

CrossPan: A Comprehensive Benchmark for Cross-Sequence Pancreas MRI Segmentation and Generalization

Linkai Peng, Cuiling Sun, Zheyuan Zhang, Wanying Dou+9

Department of Radiology, Northwestern University, Chicago, USA. · Department of Electrical and Computer Engineering, Northwestern University, Evanston, USA. · Department of Computer Science, Northwestern University, Evanston, USA.+4

Image Segmentation Magnetic Resonance Imaging Medical Image Segmentation 3D Medical Image Segmentation Medical Image Benchmarks

Open paper AskPDF arXiv

Automatic pancreas segmentation is fundamental to abdominal MRI analysis, yet deep learning models trained on one MRI sequence often fail catastrophically when applied to another-a challenge that has received little systematic investigation. We introduce CrossPan, a multi-institutional benchmark comprising 1,386 3D scans across three routinely acquired sequences (T1-weighted, T2-weighted, and Out-of-Phase) from eight centers. Our experiments reveal three key findings. First, cross-sequence domain shifts are far more severe than cross-center variability: models achieving Dice scores above 0.85 in-domain collapse to near-zero (<0.02) when transferred across sequences. Second, state-of-the-art domain generalization methods provide negligible benefit under these physics-driven contrast inversions, whereas foundation models like MedSAM2 maintain moderate zero-shot performance through contrast-invariant shape priors. Third, semi-supervised learning offers gains only under stable intensity distributions and becomes unstable on sequences with high intra-organ variability. These results establish cross-sequence generalization-not model architecture or center diversity-as the primary barrier to clinically deployable pancreas MRI segmentation. Dataset and code are available at https://crosspan.netlify.app/.

Abstract

Medical Image Segmentation

Explore similar work

CrossPan: A Comprehensive Benchmark for Cross-Sequence Pancreas MRI Segmentation and Generalization

Multimodal Assessment of Pancreatic Cancer Resectability Using Deep Learning

Assessing Pancreatic Ductal Adenocarcinoma Vascular Invasion: the PDACVI Benchmark