Members

• Michael Goldwasser MailTo(Michael.Goldwasser AT SPAMFREE carle DOT com)
• Benjamin Grosser MailTo(grosser AT SPAMFREE uiuc DOT edu)
• John Hart MailTo(jch AT SPAMFREE uiuc DOT edu)
• Russ Jamison MailTo(rjamison AT SPAMFREE uiuc DOT edu)
• Amy WagonerJohnson MailTo(ajwj AT SPAMFREE uiuc DOT edu)
• Matei Stroila MailTo(stroila2 AT SPAMFREE uiuc DOT edu)
• Alex Jerez MailTo(ajerez AT SPAMFREE itg DOT uiuc DOT edu)

Announcements

• DARPA Solicitation: DUE: 2/9/07
• January Presentation of the NIH Propoal: attachment:NIHapplicationOutline
• December Presentation: PDF

Automated Design Tool

Graphical User Interface

We will use an open source and cross-platform tool: KWWidgets

Shape Modeling

New Ideas

Forget about using spherical parameterizations. It has been quite painful to realize that they don't work as it was claimed.
The most reasonable way seems to be, for now, the following: using few landmarks, build a shape space (that is almost done, Shadi is refining the method and software). Given a mandible with a defect, find the closest mandible in the shape space. Then, outline the defect on the mesh, mark does edges as boundaries, use a hole-filling algorithm (inspired from range scans algorithms) to non-rigidly transfrom the good mesh into the defected mesh. Then, subtract the two meshes: build binary 3d images and do the subtraction and some morphological operations to get the final 3d filling.

Old Ideas

• Martin Styner has some interesting research. Here are some papers:

• Bone Morphing

• Evaluation of 3D Correspondence Methods for Model Building
• A minimum description length approach to statistical shape modeling Davies, R.H.; Twining, C.J.; Cootes, T.F.; Waterton, J.C.; Taylor, C.J. Medical Imaging, IEEE Transactions on, Vol.21, Iss.5, May 2002
• A review of the MDL approach in shape modelling (for one dimensionale shapes though): Minimum Description Length Shape and Appearance Models
• Davies et al. show that we don't need manually selected landmarks. This is fantastic since Michael and me won't have to stare at the screen for hours and hours and draw landmarks, also since manual things are subjective.
• We can use PCA (principal component analysis) to build the shape space; this is a linear model of shape
• The automatic model building (choosing landmarks) is best viewed as an optimization problem. The key thing to understand is the MDL model. The Bone morphing paper suggests using a manually initialized subdivision surface (MSS). So the landmarks are obtained as follows. Start with four points on the mandibular surface, subdivide, optimize the positions of the new points by minimizing the MDL objective function, subdivde again, optimize again, and so on until the desired number of points representing the mandibular surface.
• Meanwhile, I got another good reference for the shape modelling: 3D Active Shape Models Using Gradient Descent Optimization of Description Length. According to the first author, Tobias Heimann, the implementation is almost ready to be checked in ITK.
• I have the code for the automatic landmark positioning. It remains to do a prior mesh clean-up, the code works only with genu 0 meshes. I have to implement or find something like: Removing excess topology from isosurfaces

Haptics

Papers

• A surgical simulator for planning and performing repair of cleft lips by Stephen Schendel, Kevin Montgomery, Andrea Sorokin and Giancarlo Lionetti in Journal of Cranio-Maxillofacial Surgery (2005) 33, 223-228.

Atlas of the Skull / Database

Main steps

• build initial parameterizations: split the bone in two halfs, build unit disk params using Floater mean value weights, map one of them to the outside, map them to half spheres, join them, and then run aonther optimization (Matei)
• build the data structure for generation of new params from the initial params (Shadi)
• do the optimization (find eigen stuff, define the objective function) (Mahsa)

Old stuff

We need a good number of high quality anonymized CT scans of the normal skull. Could we get this from Carle or Beckman? This seems to be solved for the moment.

Steps for building the atlas

• build a template mesh: the segmentation of an average 3D volumetric data jaw (see Statistical shape model generation using nonrigid deformation of a template mesh)
• morph the template mesh to the individual cases (we need feature correspondences, use MDL?)
• PCA of all vertices on all cases

Back to modeling

• given the defected jaw, start putting points on the good part, bring the template mesh as close as possible to these points using an optimization procedure in the PCA space

• A good reference for this: The space of human body shapes: reconstruction and parameterization from range scans

• Another paper that might help, although our atlas data does not have that much geometry variation: Cross-parameterization and compatible remeshing of 3D model

Relevant Papers

[Maxime Berar, Michel Desvignes, G. Bailly, Yohan Payan:
3D Meshes Registration: Application to Statistical Skull Model|http://www.icp.inpg.fr/ICP/publis/synthese/_mb/mb_ICIAR04.pdf]

Scaffold Design Optimization

Relevant Papers

• Reconstruction of mandibular dysplasia using a statistical 3D shape model from the project MandibleRecon at ZIB, same place where Amira is being developed.

• Design and fabrication of custom mandible titanium tray based on rapid prototyping

Software

Amira Demo Downloads – Amira is a great program for working with 3-D datasets (i.e. volumetric CT, MRI, etc.)

Grants

Collaborations with National Centers for Biomedical Computing

National Alliance for Medical Image Computing

Meeting Schedule

Each table cell represents one hour starting at the time specified in the row heading.
Please, mark with an "x" what is not convenient for you: