| |
Discussion and Future Work
187
·
Quality measures can be used to evaluate results if a good gold standard (reference)
exists. The gold standard can be created by an experienced human user by changing
(correcting) results that were originally produced by mrFindBorders (possible within
mrLoadRet). Reference results can also be newly created from scratch. Latter way
causes a problem because sometimes it is not clear how to decide because of noise.
For example, if a certain pixel can belong either to V1 or to V2d – because there is not
enough information to make an unambiguous decision, even in case a human expert
evaluates the data – and the algorithm has decided for either of the two possibilities,
this decision should not be classified as wrong. Therefore, it is better to correct a result
by mrFindBorders instead of creating reference results from scratch. Corrections
should then only be made where the situation is clear and not where the decision is
ambiguous.
·
It is possible to use different settings to generate results that can be evaluated with a
measure of quality. There are many parameters in mrFindBorders that can be varied,
e.g., different numbers of control points, different coherence thresholds, different en-
ergy weights, etc. Different matching algorithms can also be evaluated as well as dif-
ferent ways to pre-process data.
8.7 User Interface
·
Noisy regions can be excluded12 in the measured phase-wedge image before the
matching procedure is started. The program can easily be extended so that this func-
tionality is also provided for the measured phase-ring image.
·
It is possible to deform an automatically deformed atlas manually to improve the result
provided by mrFindBorders. However, the current implementation only allows assign-
ing displacement values to control points. It is therefore difficult to shift a boundary. It
would be helpful to be able to drag a boundary in a wanted position. Then, the atlas
12
By setting regions in the measured phase-wedge image to NaN.
|  |
|
| |
|
|
