RIDER-LungCT-Seg

Manual primary gross-tumor-volume (GTV) segmentations of non-small-cell lung cancer (NSCLC) on same-day test–retest chest CT, paired with their source CT images, from RIDER-LungCT-Seg on The Cancer Imaging Archive (segmentations from Aerts et al., Nat. Commun. 2014; base images from the RIDER "Coffee-break" Lung CT collection, Zhao et al.).

This mirror is a precisely-isolated slice of a shared TCIA collection — read carefully. On TCIA, the base "RIDER Lung CT" collection (32 patients, 2072 series) pools three analysis results under one name: RIDER-LungCT-Seg (these GTV segmentations), QIBA-VolCT-1B (volumetric-measurement SR/PR/SEG objects), and a QIN CT challenge (algorithm-output SEGs). This repository contains only RIDER-LungCT-Seg — the 59 QIICR DICOM SEG + 59 Varian RTSTRUCT GTV objects — plus the 59 base CT series they reference. The QIBA / QIN objects and the redundant CT reconstructions (the full collection has 561 CT series / ~43 GB of multi-parameter reconstructions) are excluded.

Dataset Details

Field	Value
Modality	CT (chest) + DICOM SEG + RTSTRUCT
Body part	Lung / thorax — NSCLC primary tumor
Task	3D primary-GTV (tumor) segmentation
Patients	31 (RIDER-#########)
GTV objects (patient × timepoint)	59 = 30 TEST + 29 RETEST
Test–retest structure	each patient scanned twice the same day (~15 min apart); 28 patients have both timepoints, 3 have a single timepoint (2 TEST-only, 1 RETEST-only)
CT series	59 (one referenced volume per GTV object), 14,319 DICOM slices
Mask formats	DICOM SEG (QIICR, pre-rasterized) and RTSTRUCT (Varian Eclipse, source contours) — both provided for every object
Total size	~8 GB
License	CC BY 3.0

Annotation tiers & recommended ground truth

Every mask object bundles two tiers of the primary GTV:

Tier	DICOM SEG	RTSTRUCT ROI	Role
Manual (gold standard)	SegmentAlgorithmType = MANUAL	GTVp_{test,retest}_man	radiation-oncologist contour — evaluation reference
Automatic	SegmentAlgorithmType = AUTOMATIC	GTVp_{test,retest}_auto	in-house autosegmentation (method under study)

Recommended gold-standard ground truth: the MANUAL segment. The manual GTV is present for all 59 objects; the automatic GTV is present for 58 (one RETEST case, ('GTVp_retest_man',), is manual-only). For each object the manual segment number is recorded in series_index.json as manual_segment_number (typically segment 1 in the DICOM SEG); the automatic tier is auto_segment_number.

Cross-dataset overlap

• LIDC-IDRI / LUNA16: no overlap. RIDER Lung CT is a 32-patient MSKCC NSCLC test–retest cohort, distinct from LIDC sites. (The RIDER↔LIDC entanglement people warn about belongs to the separate "RIDER Pilot" collection, not this one.)
• QIBA-VolCT-1B: ⚠️ reuses the identical base CT series (same 32 RIDER Lung CT patients) and adds measurement objects. If you ever co-ingest QIBA-VolCT-1B, deduplicate by SeriesInstanceUID. The CT SeriesInstanceUID and PatientID are preserved in the folder names and in series_index.json.

Structure

images/<PatientID>/<TEST|RETEST>/<ct_SeriesInstanceUID>/*.dcm    # base CT volume
seg/<PatientID>/<TEST|RETEST>/<seg_SeriesInstanceUID>/*.dcm      # DICOM SEG (2 segments: manual + auto)
rtstruct/<PatientID>/<TEST|RETEST>/<rt_SeriesInstanceUID>/*.dcm  # RTSTRUCT (ROIs GTVp_*_man / _auto)
series_index.json

series_index.json has one record per (patient_id, timepoint) with: the CT / SEG / RTSTRUCT SeriesInstanceUIDs and relative paths, study_date, ct_num_slices, the full SEG segments table, manual_segment_number / auto_segment_number, and the RTSTRUCT roi_names. Each DICOM SEG references its source CT via ReferencedSeriesSequence; each RTSTRUCT via ReferencedFrameOfReferenceSequence.

Source & Citation

• RIDER-LungCT-Seg (segmentations): https://www.cancerimagingarchive.net/analysis-result/rider-lungct-seg/ — DOI 10.7937/tcia.2020.jit9grk8
• RIDER Lung CT (base images): https://www.cancerimagingarchive.net/collection/rider-lung-ct/ — DOI 10.7937/k9/tcia.2015.u1x8a5nr
• Official, author-deposited (MSKCC / TCIA); fully public, CC BY 3.0, no registration.

@article{aerts2014radiomics,
  author  = {Aerts, Hugo J. W. L. and Velazquez, Emmanuel Rios and Leijenaar, Ralph T. H. and others},
  title   = {Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach},
  journal = {Nature Communications},
  volume  = {5},
  pages   = {4006},
  year    = {2014},
  doi     = {10.1038/ncomms5006}
}

@article{zhao2009reproducibility,
  author  = {Zhao, Binsheng and James, Lawrence P. and Moskowitz, Chaya S. and others},
  title   = {Evaluating Variability in Tumor Measurements from Same-day Repeat CT Scans of Patients with Non-Small Cell Lung Cancer},
  journal = {Radiology},
  volume  = {252},
  number  = {1},
  pages   = {263--272},
  year    = {2009},
  doi     = {10.1148/radiol.2522081593}
}

@misc{zhao2015riderlungct,
  author    = {Zhao, Binsheng and Schwartz, Lawrence H. and Kris, Mark G. and Riely, Gregory J.},
  title     = {Coffee-break lung CT collection with scan images reconstructed at multiple imaging parameters (RIDER Lung CT) [Data set]},
  year      = {2015},
  publisher = {The Cancer Imaging Archive},
  doi       = {10.7937/k9/tcia.2015.u1x8a5nr}
}

@article{clark2013tcia,
  author  = {Clark, Kenneth and Vendt, Bruce and Smith, Kirk and others},
  title   = {The Cancer Imaging Archive (TCIA): Maintaining and Operating a Public Information Repository},
  journal = {Journal of Digital Imaging},
  volume  = {26},
  number  = {6},
  pages   = {1045--1057},
  year    = {2013},
  doi     = {10.1007/s10278-013-9622-7}
}