Dear Authors,

I've very interested in trying out your method. However, execution is halting with an error.

My command:
./run_hallucination_v13b.sh isonew.nii,p1edit.nii tt.nii regression 5E4 100 0.01 3x3x3 24 0 yes 420,350 v_synth_t2.nii ~/MCR/bin/glnxa64/v81

Code executes until some Slice (not always same slice)

e.g.
...
Slice 53 : number of patches = 49952.
Generating dictionaries..Finding nearest neighbors..Hallucinating..CVX failed !! count = 29199.
Error using image_hallucination_v13b>(parfor body) (line 823)
In an assignment A(I) = B, the number of elements in B and I must be the same.

Error in image_hallucination_v13b (line 778)
Error in run_hallucination_v13b (line 191)
Caused by:
In an assignment A(I) = B, the number of elements in B and I must be the same.
MATLAB:index_assign_element_count_mismatch


Thank you!
Also is this code compiled, or can I see the code?

Josh

The WM peak intensity should not be the max intensity. It must be from the WM region.

If your images are skull-stripped and inhomogeneity corrected, then you can use 0. The WM peak intensity will automatically be figured out by a KDE on the histogram.

You command should be this,
./run_hallucination_v13b.sh isonew.nii,p1edit.nii tt.nii regression 5E4 100 0.01 3x3x3 24 0 yes 420 350 v_synth_t2.nii ~/MCR/bin/glnxa64/v81

The WM peaks of atlas and subject should be separated by space.

Thanks for you help, I appreciate it.
As you say, once I adjusted the WM peak parameters using the max intensity of each atlas image the algorithm completed without error. Is that the correct procedure? Or do I need to ensure my peak intensity is from WM, not from skull regions??

Q2: Is 1 atlas pair usually sufficient for good results?
Q3: Is there a way to save the training to use later on new data, or must the training occur again the next time I employ the tool?

Thanks,

FYI: In my first test case, the orientation of the output image changed. I am re-running with a larger set of data and will update if orientation remains an issue.

Thanks again,
Joshua

Josh,
This is probably happening because of an error with the WM Peak normalization. Can you post the full output ?

Thanks,
Aaron