LUNAxLIDC — 2D Slice-Based Lung CT Dataset with Nodule Masks

This dataset sits at the intersection of two complementary resources:

• LUNA16 (LUng Nodule Analysis 2016) provides the scan selection and volume IDs. LUNA16 is a curated subset of LIDC-IDRI filtered to scans with slice thickness ≤2.5mm and a single reconstruction kernel per scan — 888 CT volumes in total. The volume_id field in this dataset corresponds directly to the LUNA16 series UIDs.

• LIDC-IDRI (Lung Image Database Consortium and Image Database Resource Initiative) provides the nodule annotations. For each scan, four experienced thoracic radiologists independently marked and rated pulmonary nodules across 1,018 CT scans. Nodule masks and malignancy scores in this dataset are derived entirely from the LIDC-IDRI annotation layer, accessed via pylidc.

Here we provide a 2D slice-based version of the CT volumes, extracted alongside consensus nodule masks and malignancy scores derived from multi-radiologist agreement.

---

📦 Dataset Structure

Each entry corresponds to a single 2D axial slice from a 3D CT volume:

Field	Description
volume_id	Unique identifier for the CT scan (LUNA16 series UID)
slice_id	Index of the axial slice within the volume
image	2D CT slice, normalized to 8-bit grayscale (PNG)
cancer_mask	Binary nodule mask for the slice (PNG), derived from radiologist consensus
has_nodule	Boolean — whether any nodule is present on this slice
malignancy_score	Mean radiologist malignancy rating (1–5 scale) at the most suspicious voxel; null if no nodule on slice

---

⚙️ Preprocessing

• Only CT volumes present in the LUNA16 subset list were used
• Volumes were loaded from .mhd/.raw format using SimpleITK
• Nodule annotations were retrieved via pylidc and merged using consensus masking at 50% radiologist agreement (clevel=0.5) — a voxel is marked as nodule only if ≥2 of the 4 radiologists agreed
• Consensus masks were axis-corrected from pylidc's native (y, x, z) order to the volume's (z, y, x) order prior to projection
• Malignancy scores were averaged across radiologists per nodule cluster; the maximum score was taken voxel-wise in regions of overlapping nodules
• CT intensities were clipped to the lung window (−1000 to +300 HU) and linearly rescaled to 0–255
• All slices (nodule-positive and nodule-negative) were exported; use has_nodule to filter or balance

---

🚀 Usage

from datasets import load_dataset
import matplotlib.pyplot as plt
import numpy as np

ds = load_dataset("chehablab/LUNAxLIDC", split="train")

# Show a nodule-positive slice alongside its mask
sample = next(s for s in ds if s["has_nodule"])

fig, axes = plt.subplots(1, 2, figsize=(10, 5))
axes[0].imshow(sample["image"], cmap="gray")
axes[0].set_title(
    f"Volume {sample['volume_id']} | Slice {sample['slice_id']}\n"
    f"Malignancy: {sample['malignancy_score']:.2f}"
)
axes[0].axis("off")

axes[1].imshow(sample["image"], cmap="gray")
axes[1].imshow(np.array(sample["cancer_mask"]), cmap="Reds", alpha=0.4)
axes[1].set_title("CT Slice + Nodule Mask Overlay")
axes[1].axis("off")

plt.tight_layout()
plt.show()

---

🏷️ Malignancy Scale

Radiologists rated each nodule on a 1–5 scale per the LIDC annotation protocol:

Score	Interpretation
1	Highly unlikely to be malignant
2	Unlikely to be malignant
3	Indeterminate
4	Moderately suspicious for malignancy
5	Highly suspicious for malignancy

The malignancy_score in this dataset is the mean across all radiologists who rated a given nodule cluster (1–4 raters). Nodules with no malignancy rating were ignored.

---

⚠️ Important Notes

• Class imbalance: The majority of slices are nodule-negative. It is strongly recommended to use weighted sampling or explicit filtering when training classification or segmentation models.
• Volume-level splits: When creating train/val/test splits, always split at the volume_id level — never at the slice level — to prevent data leakage between sets.
• Nodule size: Only nodules ≥3mm in diameter are included, consistent with the LUNA16 protocol.

---

📚 Citation

If you use this dataset, please acknowledge Chehab Lab and cite the original LIDC-IDRI and LUNA16 sources:

@article{Armato2011,
  author  = {Armato, Samuel G. and McLennan, Geoffrey and Bidaut, Luc and others},
  title   = {The Lung Image Database Consortium (LIDC) and Image Database Resource Initiative (IDRI): A Completed Reference Database of Lung Nodules on CT Scans},
  journal = {Medical Physics},
  year    = {2011},
  volume  = {38},
  number  = {2},
  pages   = {915--931},
  doi     = {10.1118/1.3528204}
}

@article{Setio2017,
  author  = {Setio, Arnaud Arindra Adiyoso and Traverso, Alberto and de Bel, Thomas and others},
  title   = {Validation, comparison, and combination of algorithms for automatic detection of pulmonary nodules in computed tomography images: The LUNA16 challenge},
  journal = {Medical Image Analysis},
  year    = {2017},
  volume  = {42},
  pages   = {1--13},
  doi     = {10.1016/j.media.2017.06.015}
}

---

📜 License

This dataset is released under the Creative Commons Attribution 3.0 Unported (CC BY 3.0) license.
You may copy, modify, distribute, and use the data, even for commercial purposes, provided that appropriate credit is given to the original authors.

Chehab Lab @ 2026