---
license: cc-by-nc-4.0
tags:
- medical
- mri
- pancreas
- segmentation
- classification
- ipmn
- federated-learning
size_categories:
- 1K<n<10K
---

# Cyst-X: A Multi-Center Pancreas MRI Dataset for IPMN Segmentation and Malignancy-Risk Stratification

[![Paper](https://img.shields.io/badge/Paper-arXiv%3A2507.22017-B31B1B.svg)](https://arxiv.org/abs/2507.22017)
[![Code](https://img.shields.io/badge/GitHub-Code%20%26%20Models-black.svg?logo=github)](https://github.com/NUBagciLab/Cyst-X/)
[![OSF](https://img.shields.io/badge/OSF-Dataset%20Mirror-007ec6.svg)](https://osf.io/74vfs/)

Pancreatic cystic neoplasms, including Intraductal Papillary Mucinous Neoplasms (IPMNs), are increasingly detected due to advancements in imaging technologies and growing awareness. Early diagnosis and accurate characterization of IPMNs are essential for guiding appropriate clinical management and reducing unnecessary interventions.

To support research in this domain, we present **Cyst-X**, the largest multi-center pancreas MRI resource to date. It comprises 723 T1-weighted (T1W) and 738 T2-weighted (T2W) abdominal MRI scans from 764 distinct patients collected across seven international clinical centers:
* New York University Langone Health (NYU)
* Mayo Clinic Florida (MCF)
* Northwestern University (NU)
* Allegheny Health Network (AHN)
* Mayo Clinic Arizona (MCA)
* Istanbul University Faculty of Medicine (IU)
* Erasmus Medical Center (EMC)

## 📁 Dataset Directory Structure

```plaintext
Cyst-X/
├── data splits/
│   ├── classification/
│   │   ├── 2-class/
│   │   │   ├── t1.csv                          # 4-fold CV splits for T1W (High-Risk vs. No/Low-Risk; stratified center-wise)
│   │   │   ├── t2.csv                          # 4-fold CV splits for T2W (High-Risk vs. No/Low-Risk; stratified center-wise)
│   │   │   └── two_inputs.csv                  # 4-fold CV splits for paired (T1W+T2W) multi-modal fusion; stratified center-wise
│   │   └── 3-class/
│   │       ├── t1.csv                          # 5-fold CV splits for T1W (No-Risk vs. Low-Risk vs. High-Risk; pooled across all centers)
│   │       └── t2.csv                          # 5-fold CV splits for T2W (No-Risk vs. Low-Risk vs. High-Risk; pooled across all centers)
│   └── pansegnet segmentation/
│       ├── Task210_PancreasT1MRI_splits.csv    # 5-fold cross-validation splits for T1W pancreas segmentation
│       └── Task211_PancreasT2MRI_splits.csv    # 5-fold cross-validation splits for T2W pancreas segmentation
├── IPMN_Classification/                        # IPMN classification dataset
│   ├── t1/                                     # Cropped 3D pancreatic ROIs for single-input T1W (.nii.gz)
│   ├── t2/                                     # Cropped 3D pancreatic ROIs for single-input T2W (.nii.gz)
│   ├── IPMN_labels_t1_total.xlsx               # Comprehensive labels and metadata for all T1W scans
│   ├── IPMN_labels_t2_total.xlsx               # Comprehensive labels and metadata for all T2W scans
│   └── IPMN_labels_total.xlsx                  # Master cohort metadata and paired (T1W+T2W) ground truth labels
└── IPMN_images_masks/                          # Pancreas segmentation dataset
    ├── t1/
    │   ├── images/                             # Full-volume T1W NIfTI scans (.nii.gz)
    │   └── masks/                              # Full-volume ground-truth pancreas masks (.nii.gz)
    └── t2/
        ├── images/                             # Full-volume T2W NIfTI scans (.nii.gz)
        └── masks/                              # Full-volume ground-truth pancreas masks (.nii.gz)
```