Hi Saad,
thank you much for your answer. To make it a bit clearer to myself, I would like to add a few questions, if you don't mind, and am looking forward to your comments:
So, practically, instead of using multiple masks, would this protocol work for me (using classification approach):
The aim: Determine what medullar voxel has what normalized probability (VCP) to be connected to M1, for instance:
1) Read values in fdt_path M1=>Med masked by Med_mask divided each by waytotal = A ( p that medullar voxel gets connection from M1)
2) Read values in seed_to_M1 in Med=>M1 to give spinal voxel probability to target back M1 mask, diveded each again by waytotal from upward tracking = B
3) P (medullar, ech voxel) = A + B - A*B ?
Will this give me the correct answers? Do I need the waytotal step during both steps?
Finally, do I have to add another normalization step. I am asking since the medullar mask of course will be the same but the cortical masks will differ in their voxel sizes, therefore sample sizes (voxel*5000). I am wondering whether I'd have to normalize the results by the amount of those M1 voxels who are 'thresholded' bidirectionally connected to Medulla (seed_to file from downward, fdt_path from upward tracking). I hope this is not completely confusing and I am not thinking to complicated.
I thank you much for your help.
Best
Robert
======================
Hi Robert,
In multiple masks mode, the fdt_paths volume will contain the sum of all the fdt_paths that you would have got if you ran each seed separately, with the other seeds grouped as a single waypoint mask.
In general, the numbers in fdt_paths have the following meaning: what are the chances of hitting each voxel in the brain if I repeated the same streamline experiment through the diffusion field with a different noise realisation? It is the histogram of a spatial probability distribution discretised into voxels.
Using the various mask options in probtrackx can then be thought of as calculating a conditional probability.
For example, using a waypoint equates to conditioning on the streamlines going through a given region.
Combining the fdt_paths from A->B and B->A can be done using simple laws of probabilities.
For example, if you are interested in the probability of repeating the A-B OR the B->A experiment at a given voxel, you can write:
p(A->B OR B->A) = p(A->B) + p(B->A) - p(A->B AND B->A)
where if you assume that the two events are independent (which is probably not true!):
p(A->B AND B->A) = p(A->B)*p(B->A)
However, before doing all this you need to normalise each of the distributions (divide by waytotal in each case).
I hope this is not too confusing,
Cheers,
Saad.
|