EVALUATION OF FEATURE-BASED REGISTRATION ALGORITHMS: Current Research

Click on image to see research conducted at the University of Pennsylvania.

Considerable progress was made during the current grant year (9/1/03-8/31/04) in the following areas: (i) data set selection and landmark identification; (ii) creation of the reprocessing pipeline, and (iii) testbed architecture and implementation. Forty high-quality 1.5T MRI brain volumes with associated fMRI time series were selected from the ICBM data set to be used for registration testing. A set of 36 unique point landmarks on the cortical surface was defined by Stephen Frey at the Montreal Neurological Institute (MNI) and independently labeled by two anatomists at the University of Minnesota (UMN) and two anatomists at the MNI. An algorithm for simulating regional cortical atrophy-- to be used for registration testing-- was developed at the University of Pennsylvania. A common preprocessing pipeline was defined for manual labeling and for providing nonlinear registration programs with automatically stripped, bias-corrected, tissue-classified, Talairach-aligned volume sets to use as input. Substantial progress was made in finalizing the specification of registration testbed software, and a significant amount of this software has been implemented at the UMN. During the next grant year we anticipate completion of the core software for the registration testbed and goodness of warp performance measures, along with full manual labeling of at least 20 T1 MRI brain volumes.