Roger P. Woods 1 Units for Reporting Registration Errors 491

UCLA School of Medicine 2 Validation by Visual Inspection 493

3 Validation of Point Fiducial-Based Registration and Cross-Validation Using

External Fiducial Gold Standards 493

4 Cross-Validation 494

5 Sensitivity to Starting Parameters and Statistical Modeling 494

6 Simulations 495

7 Phantoms and Cadavers 495

8 Internal Consistency 495

9 Validation of Intersubject Warping 495

References 496

From the user's perspective, accuracy is one of the most important properties of a registration method. In a research setting, relative accuracy may be a basis for selecting one method over another, and in a clinical context, knowledge of absolute accuracy may be needed to make appropriate decisions. If a particular structure is of special interest, the accuracy at this particular location, as distinct from all other locations, may need to be established. To the extent that accuracy has substantial regional variations, methods used to report accuracy need to reflect these variations. Validation of registration accuracy is generally not an easy task, because the true answers (i.e., a set of gold standard answers that can serve as a basis for measuring accuracy) are generally not available. Even when estimated gold standards are available, it often turns out that uncertainty in the gold standards themselves limits the ability to assess true accuracy. In this case, strategies that at least put limits on the true accuracy are informative. Many different validation methods have been reported in the literature, and in most cases it is difficult to compare the accuracy claimed for one method with the accuracy claimed for another because of methodological incompatibilities. This chapter will discuss some of the options and strategies available for validation studies.

0 0

Post a comment