open-discussion
open-discussion > RE: Problem with SPHARM-PDM on new dataset!
Jan 2, 2017 05:01 PM | Martin Styner
RE: Problem with SPHARM-PDM on new dataset!
Dear Nazanin
You nailed one of the major issues with SPHARM correspondence, i.e. it has big problems with objects that show a high degree of symmetry (it does not need to be spherical, ellipsoidal would be enough).
Thus, the object does not necessarily need to be elongated, but it needs to be shaped such that there is little symmetry along any of the axes.
If you have an object that has such symmetry, e.g. the amygdala is a brain structure that is almond shaped and thus SPHARM often has issues when doing amygdala shape analysis, then the best would be to use additional landmarks (at least 3) that are incorporated into the correspondence. Thus rather than using the first order ellipsoid for the parameter (correspondence) space rotation, those landmarks would be used. The current distribution does not really support that, but we are incorporating exactly that into the next release (called SALT).
Btw, if you just need to flip the parametrization to account for symmetry issues, then you can do that in the current distribution with the ParaToSPHARMPDM tool.
Martin
Best regards
Martin
Originally posted by Nazanin M:
You nailed one of the major issues with SPHARM correspondence, i.e. it has big problems with objects that show a high degree of symmetry (it does not need to be spherical, ellipsoidal would be enough).
Thus, the object does not necessarily need to be elongated, but it needs to be shaped such that there is little symmetry along any of the axes.
If you have an object that has such symmetry, e.g. the amygdala is a brain structure that is almond shaped and thus SPHARM often has issues when doing amygdala shape analysis, then the best would be to use additional landmarks (at least 3) that are incorporated into the correspondence. Thus rather than using the first order ellipsoid for the parameter (correspondence) space rotation, those landmarks would be used. The current distribution does not really support that, but we are incorporating exactly that into the next release (called SALT).
Btw, if you just need to flip the parametrization to account for symmetry issues, then you can do that in the current distribution with the ParaToSPHARMPDM tool.
Martin
Best regards
Martin
Originally posted by Nazanin M:
Dear Martin,
Thanks again for your detailed and helpful explanation.
Since this correspondence is based on the alignment of first order ellipsoids, what if the shapes are not long in one axis (for example hippocampus is elongated along one axis), then how should we get the correspondence? When our shapes is spherical, the first order ellipsoid could be rotated along any direction, which does not give the correct object-inherent alignment for getting correspondent points!
What should we do in this case? Does spharm work only with elongated shapes?
I would really appreciate with your help as I'm stuck with this issue for a while.
Thanks,
Nazanin
Thanks again for your detailed and helpful explanation.
Since this correspondence is based on the alignment of first order ellipsoids, what if the shapes are not long in one axis (for example hippocampus is elongated along one axis), then how should we get the correspondence? When our shapes is spherical, the first order ellipsoid could be rotated along any direction, which does not give the correct object-inherent alignment for getting correspondent points!
What should we do in this case? Does spharm work only with elongated shapes?
I would really appreciate with your help as I'm stuck with this issue for a while.
Thanks,
Nazanin
Threaded View
Title | Author | Date |
---|---|---|
Nazanin M | Sep 27, 2016 | |
Nazanin M | Nov 4, 2016 | |
Martin Styner | Nov 10, 2016 | |
Nazanin M | Dec 26, 2016 | |
Martin Styner | Jan 2, 2017 | |
Nazanin M | Jan 2, 2017 | |
Martin Styner | Jan 5, 2017 | |
Nazanin M | Jan 11, 2017 | |
Martin Styner | Jan 17, 2017 | |
Nazanin M | Jan 17, 2017 | |
Nazanin M | Jan 17, 2017 | |
Nazanin M | Nov 2, 2016 | |
Martin Styner | Nov 4, 2016 | |
Antoine Bouyeure | Oct 30, 2016 | |
Martin Styner | Oct 31, 2016 | |
D J | Nov 1, 2016 | |
Antoine Bouyeure | Nov 1, 2016 | |
Martin Styner | Nov 1, 2016 | |
Antoine Bouyeure | Oct 29, 2016 | |
Nazanin M | Oct 30, 2016 | |
Martin Styner | Oct 31, 2016 | |