2014 Summer Project Week:Surface approximation from contour points

• 

  Projects List

• 

  Model Maker smoothed surface for prostate peripheral zone manual segmentation

• 

  Poisson reconstructed surface with normals estimated from Model Maker surface

• 

  Left:Vessel surface reconstructed with our method. Right:Vessel surface reconstructed with Model Maker

• 

  Hausdorf distance between two surface contours

Key Investigators

• Chenxi Zhang, SPL/FudanU
• Csaba Pinter, QueensU
• Andrey Fedorov, SPL

Project Description

Large spacing and inconsistencies in contours between adjacent slices can often result in unrealistic reconstructed surfaces when using interpolation, or marching cubes. We investigate one approach to remedy this issue by approximating the object surface from the contour points. This work is based on an implementation of the Poisson surface reconstruction algorithm in the VTK framework, David-Doria Arnaud-Gelas, Rensselaer Polytechnic Institute & Harvard Medical School [1].

Objective

• Develop a Slicer module for approximating object surface from point cloud.

Approach, Plan

• Initial approach is based on the following steps
  • Extract contours from a mask image
  • Get points on the contours
  • Construct a initial surface using Slicer Module of "Model Maker"
  • Look for a closest point to each contour point from "Model Maker" result
  • Assign the normal of the closest point to the corresponding contour point
  • Reconstruct a surface from contour points using Poisson surface reconstruction algorithm
• Discuss alternative approaches for estimating normals
• Evaluate applicability on other datasets

Progress

• Initial implementation as a Slicer CLI module is available
• initial evaluation is done for the prostate manual segmentation contours from this publicly available dataset

References

[1] Doria, D., and A. Gelas. "Poisson surface reconstruction for VTK." The VTK Journal (2010). URL