Collaboration/Iowa/Meshing/Migrate Iowa Neural Net code to pure ITK

2007_Programming/Project_Week_MIT

Return to Project Week Main Page

Atlas Hand Image

Subject Hand Image

Warped Atlas Image

Key Investigators

• Iowa: Vincent Magnotta and Nicole Grosland
• Kitware: Stephen Alyward

Objective

Conversion of the Iowa Neural netwok segmentation using a flexible module for image registration supporting all ITK transforms written using the transform I/O in ITK. Conversion to the ITK neural network from the current annie implementation may also be

Approach, Plan

Integrate additional ITK transform types into the neural network using the general ITK transform I/O mechanism. Evaluate the changes to segment the phalanx bones bones of the hand.

Progress

Two milestones were reached in this effort.

• The first was that we have integrated a rigid body initialization for a Thirion Demons registration for atlas <-> subject registration. The above modification was used to warp the atlas to the subject as shown in the figures.

• The second was that the neural network code now supports all ITK transforms for registration of the atlas<->subject. This was achieved using the itkTransformReader and the a templated function over the image type and transform type to resample the images.

References

• Magnotta VA, Heckel D, Andreasen NC, Cizadlo T, Corson PW, Ehrhardt JC, Yuh WT. "Measurement of brain structures with artificial neural networks: two- and three-dimensional applications", Radiology 211 (1999), 781-90.
• Powell S, Magnotta VA, Johnson HJ, Jammalamadaka VK, Andreasen NC. "Registration and Machine Learning Based Automated Segmentation of Subcortical and Cerebellar Brain Structures", NeuroImage, In Press.