PE-Positive Digital Subtraction CTPA with Annotations

This dataset is for pulmonary embolism diagnosis using non-contrast CT and CTPA.

We provide high-quality, pixel-level annotations for emboli masks on both CTPA and non-contrast CT (registered from CTPA).

Data Format

• All files are in numpy.float16 format, with a shape of [512, 512, 512].

• Resolution: [334/512 mm, 334/512 mm, 1 mm]

• DICOM attribute: Image Orientation (Patient): [1, 0, 0, 0, 1, 0]

• Segmentation Format: Binary masks

• CT Intensity Scaling: CT_data * 1600 - 600 restores the original Hounsfield Unit (HU) values

We also provide pulmonary artery-vein segmentations and emboli segmentations for the RSNA-PE Dataset.

For the RSNA-PE Dataset, first convert DICOM to rescaled_CTPA using: 🔗 get_rescaled_ct.py
Then, normalize its Patient Orientation using: 🔗 normalize_direction.py

Dataset Structure

📂 CT-CTPA-Paired_Dataset (200 cases)
├── 📁 CTPA_Data
│        ├── 📁 rescaled_CTPA
│        ├── 📁 semantics
│            ├── 📁 emboli_mask
│            ├── 📁 airway_mask
│            ├── 📁 artery_mask
│            ├── 📁 vein_mask
│            ├── 📁 blood_center_line
│            ├── 📁 blood_branch_map

├── 📁 Non-Contrast_Data
│        ├── 📁 rescaled_ct
│        ├── 📁 semantics
│            ├── 📁 emboli_mask
│            ├── 📁 blood_vessel_lumen_mask
│            ├── 📁 airway_mask
│            ├── 📁 artery_mask
│            ├── 📁 vein_mask
│            ├── 📁 blood_center_line
│            ├── 📁 blood_branch_map

📂 AV-Emboli_Seg_RSNA-PE-Dataset (7,222 cases)
├── 📁 artery_mask
├── 📁 emboli_mask
├── 📁 vein_mask