ĄMost image
registration methods make 3 algorithmic choices:
˛How
to measure mismatch E (for error)
between I(T[x]) and J(x)?
˛How
to adjust parameters of T[x] to
minimize E?
˛How
to interpolate I(T[x]) to the J(x) grid?
ĺSo can
compare voxel intensities directly
ĄExisting AFNI
programs match images by grayscale (intensity) values
˛ E
= (weighted) sum of squares differences = Sx w(x) á
{I(T[x]) - J(x)}2
ĺOnly
useful for registering Ôlike imagesŐ:
íSPGR…SPGR,
EPI…EPI, but not SPGR…EPI
˛Parameters
in T[x] are adjusted by Ňgradient
descentÓ
˛Several
interpolation methods are available:
ĺDefault
method is Fourier interpolation
ĺPolynomials
of order 1, 3, 5, 7 (linear, cubic, quintic, and heptic)
ĄAlternative
method would be to match features computed from grayscale images:
˛Brain
outline
˛Edges
(places where image intensity changes abruptly in 1-2 pixels)
˛Such
techniques can be used to match SPGR…EPI
volumes
ĺProgram
3dAnatNudge
can estimate SPGR…EPI
translations
ĺBut
not rotations or warping