EVALUATION OF FEATURE-BASED REGISTRATION ALGORITHMS: Overview
 |
| Click on image for larger view. |
This discussion was contributed by Christos Davatzikos, Dinggang Shen, and Zhong Xue from the University of Pennsylvania.
The objective of the project is to systematically evaluate and compare the performance of several fully-automated feature-based registration algorithms. Therefore, expertly-segmented brain data will be generated for the evaluation of intersubject registration, and the accuracy of registration algorithms will be evaluated. Moreover the performance of different features used for registration will also be studied based on developing a test bed which can utilize different features and image similarity measure.
Specifically, our tasks include:
- Create a general construction model for different simulations. Develop a landmark/label-based simulator of deformations, which combines interpolation schemes with our topology-correcting transformation method and provides less-biased deformations.
- Evaluate the performance of different features (attribute vectors) for defining correspondences among images and for performing image registration.
Why Explore Feature Based Registration Algorithms?
Currently, two of INC's projects (Consensus Patterns in Functional Neuroimaging and Computational Anatomy and Visualization) highlight our central focus of modeling and visualization of spatial and temporal patterns of functional activation in the living human brain. However, both of these projects require high-quality image registration in order to successfully address the research questions being investigated. While numerous feature-based 3D registration algorithms for inter- and intrasubject registration of PET, MRI, and fMRI brain volumes have been proposed, the performance of such algorithms has yet to be optimized with regard to feature hierarchy and selection. In addition, "goodness-of-warp"criteria may vary depending upon the research question being addressed or upon the type and quality of MRI/fMRI data.
To address these issues, we propose to systematically evaluate and compare the performance of several automated feature-based registration algorithms. In doing so, we plan to (i) develop a database of 20 expertly-segmented high-resolution multimodal (T1, T2, PD) whole-brain MRI volumes for the evaluation of intersubject brain image registration, (ii) systematically evaluate the accuracy of automated algorithms for intersubject brain image registration, (iii) create a software framework that allows developers of automatic registration algorithms and users of interactive registration algorithms to systematically explore the utility of using different feature maps for an algorithm's internal computation of image/brain similarity, and (iv) post our labeled brain volumes, software modules and test results on the INC Web site as downloadable distribution sets, together with documentation and log files.