Awesome Lung CT Datasets

Why This Project?

If you've ever worked on a medical imaging research project, you know the pain: finding the right public dataset is harder than it should be. Datasets are scattered across TCIA, Zenodo, Grand Challenge, Mendeley, and random university pages. Licenses vary, formats differ, and annotations come in all shapes.

I built Awesome Lung CT Datasets to solve this problem — a single, well-organized reference for every publicly available lung CT dataset with segmentation annotations. Whether you're working on nodule detection, COVID-19 lesion segmentation, airway extraction, or multi-organ segmentation, this repository gets you to the data in seconds.

What's Inside

The repository currently indexes 15+ datasets covering 100,000+ CT scans across five categories.

Scan Count Distribution

The following chart shows the number of scans available in each dataset. Note the logarithmic scale — NLST alone contains over 75,000 scans, while specialized datasets like AeroPath focus on quality with 27 carefully annotated volumes.

General Lung Segmentation

Dataset	Scans	Annotations	Format
LIDC-IDRI	1,018	Nodule annotations by 4 radiologists	DICOM + XML
LUNA16	888	Lung nodules ≥3mm	DICOM
NSCLC Radiogenomics	211	Tumor segmentations + genomic data	DICOM
Medical Decathlon Lung	96	Lung tumor segmentations	NIfTI

COVID-19 Lung CT

Dataset	Scans	Annotations	Format
COVID-19 CT Seg Dataset 1	100 slices	Ground-glass, consolidation, effusion	NIfTI
COVID-19 CT Seg Dataset 2	829 slices	COVID-19 lesions	NIfTI
COVID-19 CT Lung & Infection	20 volumes	Left/right lung + infections	NIfTI

Thoracic Organ Segmentation

Dataset	Scans	Annotations	Format
SegTHOR	60	Heart, trachea, aorta, esophagus	NIfTI
TotalSegmentator	1,228	117+ anatomical structures	NIfTI
LCTSC 2017	60	Lung, heart, spinal cord, esophagus	DICOM

Airway Segmentation

Dataset	Scans	Annotations	Format
ATM'22	500	Full airway tree + centerlines	NIfTI
AeroPath'23	27	Trachea + bronchi (challenging pathology)	NIfTI
AIIB23	285	ILD masks + airway-informed biomarkers	NIfTI

Specialized Datasets

Dataset	Scans	Annotations	Format
LUNG-PET-CT-DX	355	Tumor bounding boxes	DICOM + XML
NLST	75,000+	Low-dose screening scans	DICOM

Annotation Coverage

The datasets span multiple annotation types essential for different research tasks. Pixel-level segmentation dominates, but the collection also covers bounding boxes, multi-organ labels, airway trees with centerlines, and multi-radiologist consensus annotations.

• Pixel-level segmentation: LIDC-IDRI, COVID-19 datasets, TotalSegmentator
• Bounding boxes: LUNG-PET-CT-DX
• Multi-organ labels: SegTHOR, TotalSegmentator (117+ structures)
• Multi-radiologist consensus: LIDC-IDRI (4 expert annotations per nodule)
• Airway trees with centerlines: ATM'22

Research Impact

Based on publication counts, the datasets in this collection have been used in thousands of research papers. LIDC-IDRI alone appears in over 1,000 publications, making it the most widely used lung CT dataset in the world.

1. LIDC-IDRI — 1,000+ publications
2. LUNA16 — 500+ publications
3. Medical Decathlon — 200+ publications
4. COVID-19 Segmentation — 150+ publications

Who Is This For?

• PhD students starting a new lung imaging project and need data fast
• Researchers benchmarking their segmentation models across multiple datasets
• Engineers building medical AI products and looking for training data
• Radiologists interested in open datasets for validation studies

Contributing

The repository is open source under the MIT license. If you know of a public lung CT dataset that's missing, you can submit a pull request to add it. Contributions welcome!

---

Check out the full repository on GitHub — and star it if it helps your research!