Results

Summary statistics are presented on a case-by-case basis, with references to the published works provided below. It should be noted that an explicit comparison of model formulations and/or implementations is not provided, only numerical results of spatial accuracy measurements obtained from the reference landmark datasets. Details of the particular DIR formulations can be found in the provided references.

Please note that the list below is not exhaustive. Investigators are encouraged to submit accepted publications describing their use of the reference data via email to Dr. Castillo, in order to have results included in the postings below.

The format of the summary table is as follows:

Algorithms are identified in the far left column, with reference to the corresponding published work. For each reference data set, errors are reported as the mean (and standard deviation) 3D Euclidean magnitude distance between calculated and reference landmark positions for the set of validation landmarks. When available, magnitude errors are also shown in right-left (RL), anterior-posterior (AP), and superior-inferior (SI) component directions. All values are reported in units of millimeters.

Additional notes regarding data reporting format:

The error measurements in general are obtained in the "snap-to-voxel" fashion, such that calculated feature positions are rounded to the nearest integer voxel position for comparison against the reference. In this way, measured versus calculated registrations are equivalent for comparison (see reference 4DCTa below for more information). Those results which are not performed in this way are indicated in the references below;

Those results displayed with green background indicate measured errors at the observer uncertainty threshold; Unless otherwise indicated, spatial accuracy results refer to the full list of coordinate pairs for each case.

** Indicates analysis over (n=300) point-pair sets