Tissue mean - LSQ

[RaissaSouza]

Hello Jacob Reinhold,
I am using your python intensity-normalization code to harmonize my data (Nifit). When applying the least square normalization method, I noticed that the tissue mean only has 2 values. I think this indicates a problem in my data since the code could not identify WM, GM, CSF. Am I right? Could you clarify for me what is suppose to be saved in the tissue means values. Is there any documentation, pre-print for LSQ, I did not find on your paper?

[jcreinhold]

How are you checking that the tissue mean only has two values? Can you provide the code you're running and the output?

[RaissaSouza]

Sure this is the way I am running it:

def lsq_harmonization_fit(images, masks,outDir):
  lsq_normalizer = LeastSquaresNormalize()
  lsq_normalizer.fit(images,masks)
  lsq_normalizer.save_standard_tissue_means(outDir+"tissue_means.npy")

  
def lsq_harmonization_apply(images, masks,outDir):
  lsq_normalizer = LeastSquaresNormalize()
  lsq_normalizer.load_standard_tissue_means(outDir+"tissue_means.npy")
  normalized = [lsq_normalizer(image,mask) for image, mask in zip(images,masks)]
  return normalized

Before calling the apply I just printed the values of the tissue mean this way:

data = np.load("tissue_means.npy")
standard_scale = data[0, :]
percentiles = data[1, :]
print(standard_scale)
print(percentiles)

This is the output:

[1.00000003]
[2.4359278]

And I was actually wrong, there is one value for the scale and 1 for percentiles, and I am confused how only this 2 values are able to identify GM, WM, CSF.

[jcreinhold]

I'm guessing you looked at the Nyul code for the "standard scale" and "percentiles". LSQ works differently and doesn't use either. If you just print data, i.e., print(data) (using the variable names from your code) you'll see the three values where the first entry corresponds to CSF, the second to GM, and the third to WM.

[RaissaSouza]

You are totally right! I messed up the codes. Thanks for the clarification ;)