How bjonze/public-cardiac-ct-dataset works
This repository publishes cardiac CT segmentations centered on the left atrial appendage (LAA), with scripts for extracting complete-LAA surfaces, computing shape descriptors, and exploring those descriptors with PCA. The main user workflow is to obtain the segmentation data, select scans listed as having complete LAAs, and run the scripts through analysis. The README also documents scan-label content and visual comparisons with TotalSegmentator. The tree contains the scripts and assets, but not the downloadable data or ID lists; those links are represented from documentation, not inspected files.
Main entry points: stacom2025_extract_surfaces.py Extract LAA surfaces · stacom2025_compute_shape_descriptors.py Compute descriptors · stacom2025_explore_shape_descriptors.py PCA exploration · assets Dataset visualizations
Components 13
Dataset
- ImageCAS CT
- Segmentation data
- Scan ID lists
- Cardiac labels
Shape processing
- Extract LAA surfaces
scripts/stacom2025_extract_surfaces.py - LAA VTK surfaces
- Compute descriptors
scripts/stacom2025_compute_shape_descriptors.py - Per-scan descriptors
- Combined descriptors
Analysis and visuals
- PCA exploration
scripts/stacom2025_explore_shape_descriptors.py - Dataset visualizations
assets
Other components
- Researcher
- TotalSegmentator
Connections 13
- Researcher → ImageCAS CT: downloads CT
- Researcher → Segmentation data: downloads labels
- Segmentation data → Cardiac labels: provides
- Scan ID lists → Extract LAA surfaces: selects complete scans
- Extract LAA surfaces → Segmentation data: reads labels
- Extract LAA surfaces → LAA VTK surfaces: writes surfaces
- LAA VTK surfaces → Compute descriptors: supplies meshes
- Compute descriptors → Per-scan descriptors: writes per scan
- Compute descriptors → Combined descriptors: writes combined CSV
- Combined descriptors → PCA exploration: supplies features
- Researcher → PCA exploration: runs analysis
- Researcher → Dataset visualizations: views examples
- TotalSegmentator → Dataset visualizations: comparison shown