RibPull: Implicit Occupancy Fields and Medial Axis Extraction for CT Ribcage Scans

Accepted at SPIE Medical Imaging 2026
To be published in the conference proceedings

Contributors

• Emmanouil Nikolakakis - UC Santa Cruz
• Amine Ouasfi - Inria, Univ. Rennes, CNRS, IRISA, M2S
• Julie Digne - LIRIS - CNRS - Université Claude Bernard Lyon 1
• Razvan Marinescu - UC Santa Cruz

---

Overview

RibPull is a novel methodology that bridges computational geometry and medical imaging by utilizing implicit occupancy fields to represent CT-scanned ribcages. Our approach enables resolution-independent queries, direct medial axis extraction, and smooth morphological operations that are challenging with traditional discrete voxel-based methods.

Key Features

• Neural Occupancy Fields: Continuous 3D representations that handle sparse and noisy medical imaging data
• SDF Conversion: Transforms occupancy fields into signed distance fields for geometric analysis
• Medial Axis Extraction: Laplacian-based contraction for robust skeletonization
• Memory Efficient: Reduces storage by ~57% (from 4.2 MB to 1.8 MB per scan)
• Clinical Applications: Enables fracture detection, scoliosis assessment, and surgical planning

Method Pipeline

1. CT Scan Input → Volumetric computed tomography scan
2. RibSeg Segmentation → Binary ribcage segmentation and point cloud extraction
3. Neural Occupancy Training → SparseOcc learns implicit surface representation
4. Mesh Reconstruction → Isosurface extraction via Marching Cubes
5. Medial Axis Extraction → Laplacian-based contraction for skeleton generation

Dataset

This work uses the RibSeg dataset, which extends the RibFrac challenge dataset with:

• 20 manually annotated CT ribcage scans
• High-quality radiologist annotations
• Detailed rib labeling and anatomical centerlines

Installation

# Coming soon upon publication

Usage

# Example code will be provided upon release

Citation

If you find this work useful, please cite our paper:

@article{nikolakakis2025ribpull,
  title={RibPull: Implicit Occupancy Fields and Medial Axis Extraction for CT Ribcage Scans},
  author={Nikolakakis, Emmanouil and Ouasfi, Amine and Digne, Julie and Marinescu, Razvan},
  journal={arXiv preprint arXiv:2509.01402},
  year={2025}
}

Acknowledgments

We gratefully acknowledge:

• The authors of RibSeg for making their benchmark dataset publicly available
• The creators of the RibFrac dataset for their contributions to medical imaging research
• SparseOcc methodology by Ouasfi et al. for unsupervised occupancy learning