Recent Changes - Search:

PmWiki

pmwiki.org

edit SideBar

RohlfingCURA

Image Similarity and Tissue Overlaps as Surrogates for Image Registration Accuracy: Widely Used but Unreliable

T. Rohlfing, IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. 31, NO. 2, FEBRUARY 2012

This is one of several papers published by this author which discuss the role of smoothness and its effects on the design and evaluation of image registration methods.

Briefly, the standard methods used to assess the success of registration methods can be shown to give good apparent performance figures for algorithms which are (by design) a bad way to solve the problem. In this case the `Completely Useless Registration Tool'. The claim is made that for a large number of publications, our only knowledge of performance, and therefore the information which has led us to regard methods as `state-of-the-art' is based upon flawed evaluation methodologies.

The paper has some limitations (some are discussed), but there is probably an over emphasis on achieving point-to-point matching accuracy, whereas for medical clinical this is not the final measure of performance (e.g. we may not get precice registration at all locations but transferred tissue labels may still be reliable).

NAT 12/7/2013

Edit - History - Print - Recent Changes - Search
Page last modified on July 12, 2013, at 12:00 PM