# Motivation

1. ITK registration pipelines are highly parameterized.
2. Registration performance highly dependent on the selected parameter values.
3. Data specific magic numbers obtained empirically.

Look at your code and see how often the following magic numbers are used: {1…10}, {0,…10}, {5,10,15…}, {${\displaystyle 10^{n}}$}, {${\displaystyle 2^{n}}$}, prime numbers, 42. Most likely, these are not optimal values.

# Goal

The goal of this project to develop a framework for automated parameter tuning of ITK non-rigid registration pipelines.

Formulate this task as an optimization over pipeline parameter values:

1. Use the tight coupling between registration and segmentation of image pairs.
2. Optimized function based on comparison of the given image segmentation and that induced by the registration.

# Data

The data used in the project can be found in this MIDAS repository.

# Team

• Ziv Yaniv, PI, (Georgetown University)
• Andinet Enquobahrie (Kitware)
• Michael Grauer ( Kitware )
• Ben Fuerst (Georgetown University/TUM)

# Ongoing

1. Analysis of the ITK optimization framework - refactor ITK hierarchy to distinguish between functions that provide derivative information and those that do not.
2. XML reading with support for schema - Xerces, check license compatibility.
3. Look into the use of Condor for distributed computing.