Hi there,
> I am trying to create a Jacobian of the field using --jout option in fnirt command.
> What i observed was that the Jacobian maps generated by fnirt' ing images A-->B and B-->A are not inverse consistent.
>
> For example: when i choose a random voxel on the Jacobian map, i see that the value for that particular voxel in both cases (A-->B and B-->A) are greater than 1. Which means that the same voxel is expanding in both trials.
there are two issues here.
1. Fnirt is not guaranteed to be inverse consistent. However, inverse consistency was a large part of the evaluation we did of fnirt and the errors were quite small (~3mm for most cortical areas).
2. I think this is what you see in your case. The Jacobian map is in the space of the reference scan. When you have done A->B and look at the Jacobian in coordinate x,y,z that corresponds to that particular area in B. If you want to know what the Jacobian in the B->A warp is in that location you can not look in x,y,z, because that refers to that location in the space of A. You need to look at the Jacobian for B->A in the location x+dx(x,y,x),y+dy(x,y,z),z+dz(x,y,z) where the displacement fields are those given by the A->B warp.
Spatial stuff is confusing.
Good luck Jesper
>
> Is there a way to make it inverse consistent, so that i see expansion in one direction (A-->B) and contraction in the opposite direction(B-->A)?
>
> BW
> Jay
>
|