Cases/ImageSegmentationVariability

Contents

Synopsis
Team Members
Progress
References

Synopsis

The overall goal of this study is to assess the effects of variations in patella cartilage surface topology due to image segmentation attempts by different users on finite element model predictions.

Preliminary Proposal: Proposal.pdf (will be moved to repository and modified based on discussions)

Use Case

This use case aims at: a) taking multiple image segmentation attempts (proposed: 6) for patellofemoral joint, b) using the exact same set of tools and protocols and build surfaces, meshes and models and, c) for a given loading scenario (similar to patellofemoral joint testing for the same specimen) comparing the peak stresses on the patella cartilage.

Simulation Aims

To assess whether segmentation variability due to different users leads to significant variations in finite element model predictions.
Secondary aim: assessing effect of material model variations along with surface topology on finite element predictions.

Model

Multiple models will be developed for oks001 from several segmentation attempts.

https://simtk.org/svn/openknee/app/SegmentationVar/src

Inputs

45 degree flexion angle and 600 N quadriceps load. (proposed, may be changed).

Outputs

Time histories of patella cartilage stress behavior.

Simulation Conditions

Team Members

Snehal Chokhandre - study planning, model development, execution of simulations and post-processing, lead authorship in publications
Ahmet Erdemir - overall supervision

Progress

First attempt to manually create a simple compression model for patella and femur cartilage. Geometries generated by Craig for oks001 (smoothed) were used for this trial. Both the cartilage geometries were reoriented in Blender to overlap each other and for easier application of boundary conditions. Femur cartilage also truncated in the region not overlapped by the patella cartilage.
- Manual meshing and FEA procedure:
  - Geometries imported in Salome (stl files created from Slicer 3D, smoothed surfaces)
  - Tetrahedral meshes and node sets created in Salome. Exported as .unv files
  - Converted .unv to .inp files using unv2abaqus.py.
  - Removed surface elements from the .inp file using text editor before importing in FEBio Preview.
  - Imported .inp files in FEBio Preview, applied boundary conditions, defined materials (Mooney-Rivlin), defined contact, exported as .feb.
  - Ran simulation using FEBio and results visualized in FEBio Postview.
First segmentation and geometry generation attempt using specifications.
- LVTIT and low mesh density (6). For details check, Specifications/ImageSegmentation and Specifications/GeometryGeneration
- Hausdorff Distances for the geometry compared to attempt by Craig (using as reference geometry)
- There is a large region with >1mm variability.
- Compared segmentation with reference and image set by overlaying them in 3DSlicer.
- Identified regions where geometry boundaries were over/underestimated and corrected.
- This serve as a good example of segmentation variability due to user experience.