Annotations for the PANORAMA challenge: Public Training and Development Dataset

Imaging Dataset

To download the associated imaging data, visit the PANORAMA challenge website

The public training and development set of the PANORAMA challenge includes 80 patients without PDAC from the National Institutes of Health Clinical Center data set and 194 patients (98 PDAC) from the Medical Segmentation Decathlon pancreas dataset (Task07_Pancreas, training set). The differentiation between PDAC and non-PDAC patients in the MSD data set was derived directly from Suman, et al. (2021). Patients with a stent were removed in accordance with the PANORAMA study inclusion criteria.

Reference Standard

The public training and development set of the PANORAMA challenge includes patients with the following reference standard:

• Histopathology confirmation of pancreatic ductal adenocarcinoma (PDAC) via resection specimen, biopsy, or cytology analysis
• Histopathology confirmation of non-PDAC lesion via resection specimen, biopsy, or cytology analysis
• CT confirmation of PDAC without follow-up
• CT confirmation of non-PDAC lesion without follow-up
• CT confirmation of healthy/unremarkable pancreas without follow-up

The reference standard used for each individual can be found in clinical_information.csv

Annotations

For positive cases (PDAC) in the public training and development data set, manual lesion segmentations were generated by one of two trained investigators supervised by an expert radiologist with over 20 years of pancreatic cancer experience. Lesion deliniations were created using ITK-SNAP v3.80. There are 676 PDAC cases in the data set, of which 482 have manual annotations. For the remaining 194 cases, lesion segmentations were automatically generated using an updated version of the same model described and validated by Alves, et al (2021).

For all cases in the public training and development data set, segmentations of the pancreas parenchyma, pancreatic duct, veins, arteries, and common bile duct are provided. These segmentations were automatically generated by the algorithm described and validated by Alves, et al (2021).

Label Legend

Label	Value
Background	0
PDAC lesion	1
Veins	2
Arteries	3
Pancreas parenchyma	4
Pancreatic duct	5
Common bile duct	6

Open-Source Contributions

We encourage open-source contributions! For instance, you can contribute expert-derived delineations. If you're interested, feel free to propose PRs for inclusion to this repo. Pending quality control, substantial contributions will be merged in and credited accordingly.

Citation

If you are using this dataset or some part of it, please cite the following article:

Alves, N., Schuurmans, M., Rutkowski, D., Yakar, D., Haldorsen, I., Liedenbaum, M., Molven, A., Vendittelli, P., Litjens, G., Hermans, J., & Huisman, H. (2024). The PANORAMA Study Protocol: Pancreatic Cancer Diagnosis - Radiologists Meet AI. Zenodo. https://doi.org/10.5281/zenodo.10599559