Lung-Fused-CT-Pathology — CT + segmentation masks

Radiology side of the TCIA collection Lung-Fused-CT-Pathology (Fused Radiology-Pathology Lung), reformatted to co-registered NIfTI for the MedSphere / EasyMedSeg segmentation suite. 6 patients with surgically-resected ground-glass pulmonary nodules (lung adenocarcinoma); pre-operative chest CT was co-registered with ex-vivo whole-slide histopathology to project the true extent of invasive adenocarcinoma onto in-vivo CT.

Modality: chest CT · Organ: lung (nodule / adenocarcinoma) · n = 6 · License: CC BY 3.0

⚠️ Scope of this mirror (read first)

This repository contains the CT volumes and all CT-space segmentation masks only. The original collection also ships 21.3 GB of whole-slide histopathology images (H&E .tiff, served via Aspera) — those raw pathology slides are NOT included here, because they are not CT segmentation data and are not consumable by a CT dataloader. They remain available from TCIA (see Provenance below).

Crucially, the pathology information is not lost: the three fusion_* masks are the pathologist's annotations (lesion / invasive-adenocarcinoma component / blood vessels) mapped from the 3-D histology reconstruction onto the CT grid, so the radiology-pathology fusion that defines this dataset is preserved in CT space.

Contents

One folder per patient (LungFCP-01-0001 … -0006), each with 1 CT + 5 masks, all identical geometry (same shape / spacing / origin / direction as the CT):

File suffix	Source	Encoding	Role
_ct.nii.gz	CT	int16 HU	Chest CT volume
_nodule.nii.gz	DICOM SEG	uint8 {0,1}	Gold-standard nodule — majority vote of 3 radiologists
_vessel.nii.gz	DICOM SEG	uint8 {0,1}	Blood vessel, manual (used for validation)
_fusion_lesion.nii.gz	DICOM FUSION	uint8 {0,1}	Lesion mapped from 3-D histology (orig code 7)
_fusion_invasion.nii.gz	DICOM FUSION	uint8 {0,1}	Invasive-adenocarcinoma component mapped from histology (orig code 306)
_fusion_vessel.nii.gz	DICOM FUSION	uint8 {0,1}	Blood vessels mapped from histology (orig code 32)

metadata.csv lists, per patient, the CT reconstruction used, shape, spacing, origin, and the voxel count + original (pre-binarization) code value of each mask.

Recommended ground truth

For nodule/lesion segmentation benchmarking, use _nodule — the paper's designated gold standard ("a voxel within the nodule annotation of at least two of three raters was included in the consensus delineation"). The fusion_* masks are derived (histology-registered) ground truth of disease extent and are best treated as a distinct, harder task. Masks were binarized from the source label codes.

Per-patient CT reconstruction

The masks were drawn on one specific CT reconstruction per patient (the kernel varies — e.g. patient 0005's masks align to the lung kernel, not soft-tissue); each patient's stored CT is the recon whose voxel grid matches its masks exactly.

Patient	CT reconstruction	Shape (z,y,x)	Spacing (x,y,z) mm
LungFCP-01-0001	Thorax 2.0 B31f	(321, 512, 512)	(0.76, 0.76, 1.0)
LungFCP-01-0002	(axial, 337 sl.)	(337, 512, 512)	(0.66, 0.66, 1.0)
LungFCP-01-0003	Chest/Abd/Pel 5.0 B30f Soft Tiss	(125, 512, 512)	(0.732, 0.732, 5.0)
LungFCP-01-0004	Chest 2.0 B30f Soft Tissue	(308, 512, 512)	(0.719, 0.719, 1.0)
LungFCP-01-0005	Thorax 2.0 B70f Lung	(126, 512, 512)	(0.574, 0.574, 2.0)
LungFCP-01-0006	Lung	(385, 512, 512)	(0.801, 0.801, 1.0)

Integrity notes

• Provenance — official. Downloaded from the TCIA NBIA REST API (public, no login). Source: Madabhushi & Rusu (2018), data from University Hospitals Cleveland Medical Center + Case Western Reserve University. Patient count 6/6 matches the paper. Masks/CT are author-generated DICOM image stacks (Manufacturer = github.com/mirabelarusu/RadPathFusionLung), converted to NIfTI with SimpleITK.
• Faithful naming. This is the CT + masks subset of "Lung-Fused-CT-Pathology"; the raw whole-slide pathology component is excluded (see Scope above).
• Cross-dataset overlap — none known. Single-institution surgical cohort (Cleveland/CWRU), selected for resected GGNs with matched ex-vivo histology — disjoint from LIDC-IDRI / LUNA16 / NSCLC-Radiomics / 4D-Lung / MSD-Lung. No shared identifier column exists.

Provenance & citation

• Collection: https://www.cancerimagingarchive.net/collection/lung-fused-ct-pathology/
• Whole-slide pathology (not mirrored here): available from TCIA via Aspera / PathDB.
• Data DOI: Madabhushi, A., & Rusu, M. (2018). Fused Radiology-Pathology Lung (Lung-Fused-CT-Pathology) [Data set]. The Cancer Imaging Archive. https://doi.org/10.7937/k9/tcia.2018.smt36lpn
• Paper: Rusu, M., et al. (2017). Co-registration of pre-operative CT with ex vivo surgically excised ground glass nodules to define spatial extent of invasive adenocarcinoma on in vivo imaging: a proof-of-concept study. European Radiology, 27(10), 4209–4217. https://doi.org/10.1007/s00330-017-4813-0
• Code (fusion pipeline): https://github.com/mirabelarusu/RadPathFusionLung

License CC BY 3.0 — please cite the paper and the TCIA collection DOI. For research and educational use; not for clinical decision-making.