CraniumPy

• Description
• Usage
  • Getting started
  • Rigid registration and mesh harmonization
  • Automated measurement extraction
  • Facial asymmetry calculation
  • Non-rigid ICP
• Download CraniumPy (.exe)
• Clone repository
• Citation
• Author

Description

CraniumPy is a user-friendly tool developed for the registration and craniofacial analysis of 3D images (.ply, .obj, .stl). It can serve as an end-to-end pipeline, transforming a raw 3D image into a harmonized mesh, ready for either cranial or facial analysis. The integrated pipeline incorporates:

• Rigid and non-rigid registration.
• Voronoi-based clustering for mesh resampling.
• Artifact detection and repair algorithms.
• A validated algorithm to automatically extract commonly used head measurements.
• Functionality to quantify facial asymmetry.

This repository makes use of some excellent features implemented in the PyVista library (https://docs.pyvista.org/) and PyMeshFix (https://pymeshfix.pyvista.org/).

Usage

Getting started

Refer to the step-by-step guide to get started.

Rigid registration and mesh harmonization

Mesh registration and harmonization in CraniumPy requires three steps: 1. Landmark selection: Three anatomical landmarks (Nasion, LH tragus, RH tragus) need to be located by the user for registration.

1. Register for cranial/facial analysis: The mesh is rigidly aligned to an average normal template by computing the Euler angles between two sets of vectors (derived from placed landmarks). The registered mesh (filenamerg.ply) and the landmark coordinate positions (filenamelandmarks.json) are stored in place.

2. "Clip, Repair, Resample": In the final pre-processing step, the mesh is clipped through a reference plane (cranial analysis: through nasion-tragus plane | facial analysis: through centroid of triangle formed by the 3 landmarks). The clipped mesh is then inspected and artifacts are automatically repaired. Finally, the mesh is resampled to ensure a predefined number of vertices (default: n= 10.000).

Based on a single transverse slice (at maximum head depth), cephalometric measurements are automatically extracted and plotted on the 3D model.

The example mesh resources/test_mesh/test_mesh.ply can be used to explore the functionality of this tool.

NOTE: Step 3. (Clip, Repair, Resample) is computationally heavy (due to the resampling step) and can take a while depending on your setup. Clicking the screen may cause it to freeze.

Automated measurement extraction

From a mesh that is pre-processed for cranial analysis, several cephalometric measurements can be extracted. These include: - Occipitofrontal diameter (OFD) or head depth - Biparietal diameter (BPD) or head breadth - Cephalic index (CI) - Occipitofrontal circumference (OFC) or head circumference - Mesh volume above the nasion-tragus plane (for potential intracranial volume approximation).

This algorithm is located under the tab Compute>Cephalometrics. The axial slice from which these measurements have been obtained can also be plotted separately using Compute>2D slice.

NOTE: CraniumPy has been optimized for pediatric analysis. For this reason, some hard-coded boundary conditions are established in the extraction algorithm (e.g. code line 95). If CraniumPy does not work as desired for your research application (e.g. prematures or adults), feel free to contact me to see if we can find a solution.

Facial asymmetry calculation

A mesh registered for facial analysis can be used to compute the mean distance between each vertex on one half of the face to its corresponding vertex on the other half (using a mirrored reflection). The output is a heatmap (in mm), showing which areas are more or less symmetric. A quantitative metric is also computed, the mean facial asymmetry (MFA) index, which encapsulates the overall asymmetry observed in the face. This algorithm is located under the tab compute>Evaluate Asymmetry.

Non-rigid ICP

A non-rigid iterative closest point (NICP) algorithm has been implemented which tries to deform a template onto the user's input mesh. This feature can come in handy when there's a need for point-to-point correspondence (e.g. automated landmark detection, advanced shape analysis, geometric deep learning). This step allows a user to match the topology of a mesh or set of meshes to that of the template. The mesh quality may be (slightly) reduced after NICP.

Download CraniumPy

If you want to run CraniumPy locally (on Windows) follow the steps below. You do not need to install any requirements or dependencies.

1. Download the latest version: CraniumPy (v0.4.2)

2. Extract the .zip folder in any location

   The extracted folder contains the templates (for visualizing the registration) and an example mesh is provided at `testmesh/testmesh.ply` for experimentation.

3. Run CraniumPyv0.4.2.exe_ (give it a few seconds to load)

IMPORTANT: the reference frame has been changed from v0.3.0 onwards. The new convention is based on the more commonly used right hand coordinate system in which the z-axis points outward from the nose.

Clone repository

The implemented NICP and facial asymmetry algorithm requires the scikit-sparse library. Follow these instructions to install the required dependencies: 1. Create and/or load a virtual environment (for more information about virtual environments, check the docs):

conda create -n CraniumPy python=3.8 conda activate CraniumPy

1. Clone repository: git clone https://github.com/T-AbdelAlim/CraniumPy.git cd CraniumPy
2. Install requirements: pip install -r requirements.txt

5.Install NICP dependencies (tested on Windows 10 with Anaconda 3 and Python 3.8)

• conda install -c conda-forge cython - tested with v0.29.32
• conda install -c conda-forge suitesparse - tested with v5.4.0
• conda install -c conda-forge numpy - tested with v1.23.2
• conda install -c conda-forge scipy - tested with v1.9.1
• Download Microsoft Build Tools for C++ from https://visualstudio.microsoft.com/visual-cpp-build-tools/ (tested with 2022, should work with 2015 or newer)

1. Install Visual Studio Build Tools

   1. Choose Workloads
   2. Check "Desktop development with C++"
   3. Keep standard settings

2. Run in a Powershell

   • $env:SUITESPARSE_INCLUDE_DIR='.../anaconda3/envs/CraniumPy/Library/include/suitesparse'
   • $env:SUITESPARSE_LIBRARY_DIR='.../anaconda3//envs/CraniumPy/Library/lib'
   • pip install scikit-sparse

3. Test from sksparse.cholmod import cholesky

4. Run CraniumPy: python CraniumPy.py

Known issues

• CraniumPy closes if less than 3 landmarks are selected before registration. Steps need to be followed carefully as described in the step-by-step guide
• Registration step 3. (Clip, Repair, Resample) takes a few seconds and freezes the user interface until all the calculations are performed, which may create the suspicion that the programme crashed ("windows not responding"). This is not the case. The user interface will unfreeze after the calculations are completed and will return the final mesh.

Citation

Kindly consider citing the software if it supports your research:

Initial Publication:

Abdel-Alim et al. - Sagittal Craniosynostosis: Comparing Surgical Techniques using 3D Photogrammetry. Plastic and Reconstructive Surgery ():10.1097/PRS.0000000000010441, March 22, 2023. | DOI: 10.1097/PRS.0000000000010441

Validation Study:

Abdel-Alim et al. - Reliability and Agreement of Automated Head Measurements From 3-Dimensional Photogrammetry in Young Children. The Journal of Craniofacial Surgery 34(6):p 1629-1634, September 2023. | DOI: 10.1097/SCS.0000000000009448

Github Repository: Abdel-Alim, T. (2022). CraniumPy [Computer software]. https://doi.org/10.5281/zenodo.5634153

Author

Tareq Abdel-Alim | Departments of Neurosurgery and Radiology, Erasmus MC, Rotterdam, the Netherlands

If you have any questions, suggestions, or problems do not hesitate to contact me: t.abdelalim@erasmusmc.nl