Pancreatic-CT-CBCT-SEG

Breath-hold CT and cone-beam CT (CBCT) images with expert manual organ-at-risk (OAR) segmentations from radiation treatments of locally advanced pancreatic cancer at Memorial Sloan Kettering Cancer Center.

Dataset Details

Field	Value
Modality	CT (planning, breath-hold, contrast-enhanced) + CBCT (kV, deep-inspiration breath-hold)
Body part	Upper abdomen — gastrointestinal organs-at-risk
Task	3D multi-class segmentation (2 foreground OARs)
Patients	40 (locally advanced pancreatic cancer)
Studies	40
Series (total)	370
CT series	200 (planning CT + Aligned CT registration variants + CBCT slices)
RTSTRUCT series	130 (primary contours + dual-observer subset)
RTDOSE series	40
Images (slices)	24,246 DICOM slices
CT volumes	40 planning CTs (one per patient)
CBCT volumes	80 (two per patient: 58 limited-view 200°, 22 full-rotation 360°)
Format	DICOM (CT, CBCT, RTSTRUCT, RTDOSE)
License	CC BY 4.0

Segmentation Labels

Each RTSTRUCT contains two combined organs-at-risk:

1. Stomach + first two duodenum segments (combined)
   • Code on planning CT: SDPC
   • Code on CBCT: SDCB
2. Small bowel (remaining)
   • Code on planning CT: BSPC
   • Code on CBCT: BSCB

Some structures may be tagged "UNAPPROVED" — research-team-checked but not clinically approved.

Subsets

• Planning CT (n = 40): one breath-hold contrast-enhanced helical CT per patient, contoured by the primary radiation oncologist and reviewed/edited by two trained medical physicists.
• CBCT (n = 80): two CBCT acquisitions per patient.
  • 58 limited-view (200° rotation)
  • 22 full-rotation (360°, subsequently cropped)
• Multi-observer reliability subset (v2 addition): 10 of the 80 CBCTs have additional repeat OAR delineations from two independent observers in addition to the primary contour, enabling inter-observer agreement analysis. Reported Dice in paper: small bowel 0.82 ± 0.07, stomach + duodenum 0.86 ± 0.06.

Recommended Ground Truth

The primary clinical contours (drawn by six radiation oncologists, then reviewed and edited by two medical physicists) are the recommended ground truth for all 40 planning CTs and all 80 CBCTs. They reflect the consensus standard of care. The dual-observer repeat contours exist only to quantify variability on a 10-CBCT subset and are not a complete label set.

Splits

The original publication does not prescribe train/val/test splits.

Structure

images/<PatientID>/<StudyInstanceUID>/<SeriesInstanceUID>/*.dcm
segmentations/<PatientID>/<StudyInstanceUID>/<SeriesInstanceUID>/*.dcm
series_to_patient.json

PatientID ranges from Pancreas-CT-CB_001 to Pancreas-CT-CB_040.

The images/ tree contains all CT-modality DICOMs (planning CT, Aligned CT registration variants, and CBCT). The segmentations/ tree contains all RTSTRUCT and RTDOSE DICOMs. Each RTSTRUCT references its source CT series via ReferencedFrameOfReferenceSequence / RTReferencedSeriesSequence.

Version History

• v1 (2021-10-15, ~9.3 GB): registered/resampled CBCT scans only.
• v2 (2022-08-23, current): adds RTDOSE files, original-format CBCTs, and dual-observer repeat OAR delineations on 10 CBCTs.

Source

• TCIA collection: https://www.cancerimagingarchive.net/collection/pancreatic-ct-cbct-seg/
• TCIA wiki: https://wiki.cancerimagingarchive.net/pages/viewpage.action?pageId=93258557
• DOI: 10.7937/TCIA.ESHQ-4D90
• Released: v2 on 2022-08-23 (fully public, no registration required)

Citation

@article{hong2022pancreaticctcbctseg,
  author  = {Hong, Julian and Reyngold, Marsha and Crane, Christopher and
             Cuaron, John and Hajj, Carla and Mann, Justin and
             Zinovoy, Melissa and Yorke, Ellen and LoCastro, Eve and
             Apte, Aditya P. and Mageras, Gig},
  title   = {Breath-hold CT and cone-beam CT images with expert manual
             organ-at-risk segmentations from radiation treatments of
             locally advanced pancreatic cancer},
  journal = {Scientific Data},
  volume  = {9},
  pages   = {637},
  year    = {2022},
  doi     = {10.1038/s41597-022-01758-9}
}

@misc{pancreaticctcbctseg2021tcia,
  author    = {Hong, J. and Reyngold, M. and Crane, C. and Cuaron, J. and
               Hajj, C. and Mann, J. and Zinovoy, M. and Yorke, E. and
               LoCastro, E. and Apte, A. P. and Mageras, G.},
  title     = {Pancreatic-CT-CBCT-SEG [Dataset]},
  year      = {2021},
  publisher = {The Cancer Imaging Archive},
  doi       = {10.7937/TCIA.ESHQ-4D90}
}

@article{han2021pancreaticcbctseg,
  author  = {Han, Xiao and Hong, Julian and Reyngold, Marsha and others},
  title   = {Deep-learning-based image registration and automatic
             segmentation of organs-at-risk in cone-beam CT scans from
             high-dose radiation treatment of pancreatic cancer},
  journal = {Medical Physics},
  volume  = {48},
  number  = {6},
  pages   = {3084--3095},
  year    = {2021},
  doi     = {10.1002/mp.14906}
}