PatchChestCT: A patch-level spatial annotation dataset for nine abnormalities in chest CT

PatchChestCT provides 3D patch-level binary annotations for nine clinically significant abnormalities across 2,201 physician-reviewed chest CT volumes derived from the CT-RATE dataset (Hamamci et al., Nature Biomedical Engineering, 2026).

Each annotation is stored as a binary NumPy array (.npz) with shape (24, 12, 12), corresponding to the 3D patch grid of the preprocessed CT volume.

Annotated abnormalities:
  Arterial wall calcification · Coronary artery calcification · Pericardial effusion ·  Hiatal hernia · Lymphadenopathy · Atelectasis · Lung opacity · Consolidation · Bronchiectasis

IMPORTANT — This deposit contains ONLY annotation files (.npz). No CT images, radiology reports, or original CT-RATE metadata are included. To use these annotations, you must independently obtain the source CT volumes from
the official CT-RATE repository and agree to its Data Usage Agreement:  https://huggingface.co/datasets/ibrahimhamamci/CT-RATE

Each subdirectory is named using the original CT-RATE volume identifier (e.g., train_1003_a_1/), enabling unambiguous alignment with the source data.

For alignment code and full usage instructions, see the companion code repository:  https://github.com/SadVoxel/PatchChestCT