MedOtter/LUAD-CT-Survival

LUAD-CT-Survival (Long and Short Survival in Adenocarcinoma Lung CTs)

40 contrast-enhanced chest CT scans of lung adenocarcinoma patients from the H. Lee Moffitt Cancer Center, each with an expert-seeded primary-tumor segmentation and a binary survival-outcome label (long- vs short-survivor). This is the segmentation-ready HuggingFace mirror.

Dataset Details

Format note (reformatted variant)

The original TCIA distribution ships the CT images as DICOM and the tumor masks as uncompressed NIfTI with values 0/255. This mirror converts each DICOM series to .nii.gz with SimpleITK (voxel data, spacing, origin and direction preserved) and binarizes the masks to {0, 1}. Image↔mask geometry was verified identical (same shape / spacing / origin) before binarization; any case needing it has its mask nearest-neighbour resampled onto the image grid, so every masks/<id>.nii.gz is defined on exactly its images/<id>.nii.gz grid.

Ground truth

One tumor mask per case, produced by a semi-automatic region-growing algorithm with radiologist-placed seed points. There is a single annotation tier — it is the gold-standard reference (no multi-rater or auto-vs-expert variants).

⚠️ Cross-dataset overlap (benchmark-leakage hazard)

LUAD-CT-Survival is a TCIA analysis-result collection: its masks and survival labels are original, but the 40 CT image series are drawn from two other Moffitt TCIA collections:

• —37 cases — QIN LUNG CT (patient IDs R0xxx)
• —3 cases — LungCT-Diagnosis (QIN-LSC-0009, QIN-LSC-0014, QIN-LSC-0064)

patient_id in metadata.csv equals the original TCIA PatientID and is the cross-reference key. Exclude these patient IDs before benchmarking against QIN-LUNG-CT or LungCT-Diagnosis. (The NBIA API currently files all 40 series' Collection as "QIN LUNG CT"; the three QIN-LSC-* IDs trace to LungCT-Diagnosis per the source publication, recorded in source_collection.)

Structure

images/<patient_id>.nii.gz    # CT volume
masks/<patient_id>.nii.gz     # binary tumor mask (tumor = 1)
metadata.csv                  # patient_id, survival_label, source_collection,
                              #   num_slices, spacing_x/y/z, tumor_voxels
FeaturesWithLabels.csv        # original TCIA radiomic features + survival_label (verbatim)

Source & Citation

• —TCIA collection: Long and Short Survival in Adenocarcinoma Lung CTs (LUAD-CT-Survival), DOI 10.7937/K9/TCIA.2017.0tv7b9x1, CC BY 3.0.

@article{paul2016deepfeature,
  author  = {Paul, Rahul and Hawkins, Samuel H. and Balagurunathan, Yoganand and
             Schabath, Matthew B. and Gillies, Robert J. and Hall, Lawrence O. and
             Goldgof, Dmitry B.},
  title   = {Deep Feature Transfer Learning in Combination with Traditional
             Features Predicts Survival among Patients with Lung Adenocarcinoma},
  journal = {Tomography},
  volume  = {2},
  number  = {4},
  pages   = {388--395},
  year    = {2016},
  doi     = {10.18383/j.tom.2016.00211}
}

@misc{luadctsurvival2017tcia,
  author    = {Goldgof, D. and Hall, L. and Hawkins, S. H. and Schabath, M. B. and
               Stringfield, O. and Garcia, A. and Balagurunathan, Y. and Kim, J. and
               Eschrich, S. and Berglund, A. E. and Gatenby, R. and Gillies, R. J.},
  title     = {Long and Short Survival in Adenocarcinoma Lung CTs [Dataset]},
  year      = {2017},
  publisher = {The Cancer Imaging Archive},
  doi       = {10.7937/K9/TCIA.2017.0tv7b9x1}
}