DSC perfusion pipeline validation

servoz commented 3 weeks ago

We want the same results and the same report as with amigo

servoz commented 2 weeks ago

1 Import / data conversion to NIFTI: Contrary to the results previously observed when validating the CVR pipeline, we observe no difference between the data when the conversion is carried out in mri_conv (Mia) and in Amigo. I don't understand why, but as the differences observed during the validation of the CVR pipeline were very small anyway, we won't try to explain this difference in results at this stage (although it might be interesting to understand why we observed a difference previously, different version of the libraries? error on the validation stage, etc.).

servoz commented 2 weeks ago

2 spatial_preprocessing pipeline: As mentioned in the CVR pipeline validation, Mia uses the parameters warp.mrf, warp.cleanup, and warp.fwhm for process segmentation whereas nipype does not expose these parameters (default values are used for these parameters in nipype). In order to be able to compare the results between Mia and Amigo, we set the values of these parameters to the default values in Amigo.

Under these conditions, the spatial_preprocessing pipeline in Amigo and Mia gives the same results.

servoz commented 2 weeks ago

3 spatial_mask pipeline: We get exactly the same results with the spatial_mask pipeline in Mia and Amigo

servoz commented 2 weeks ago

4 Make_AIF: We obtain virtually the same result. For the AIF, all the points are identical except for one with a very small difference (in the baseline, for the eighth point in Mia: 1855.0 and in Amigo: 1854.8). If necessary, we can come back to this result, but for the moment this very small difference should have no effect on the final result. As far as the Score parameter (see output aif_file parameter for the Make_AIF brick) is concerned, the results are very similar (the main difference concerns the value of the score, which is an arbitrary value that can vary between the calculation in Matlab and in Python). However, the trends are respected, there is no difference in the warnings, and the voxels used to calculate the AIF are broadly the same).

servoz commented 5 days ago

5 Deconv_from_aif: A close look at the mia_processes and Amigo codes shows no significant differences in the intermediate calculations. The final results are very similar overall: To rationalise these differences, a number of static parameters have been calculated:
- CBF: Statistics on rate ([0 to 1]) differences : Minimum value: 0.00 Maximum value: 0.89 Mean: 0.01 Standard deviation: 0.02 Median: 0.00 25th percentile (1st quartile): 0.00 75th percentile (3rd quartile): 0.01 99th percentile: 0.09
- CBV: Statistics on rate ([0 to 1]) differences : Minimum value: 0.00 Maximum value: 1.50 Mean: 0.00 Standard deviation: 0.01 Median: 0.00 25th percentile (1st quartile): 0.00 75th percentile (3rd quartile): 0.00 99th percentile: 0.00
- MTT: Statistics on rate ([0 to 1]) differences : Minimum value: 0.00 Maximum value: 1.10 Mean: 0.01 Standard deviation: 0.02 Median: 0.00 25th percentile (1st quartile): 0.00 75th percentile (3rd quartile): 0.01 99th percentile: 0.09
- T0: Statistics on rate ([0 to 1]) differences : Minimum value: 0.00 Maximum value: 1.48 Mean: 0.00 Standard deviation: 0.01 Median: 0.00 25th percentile (1st quartile): 0.00 75th percentile (3rd quartile): 0.00 99th percentile: 0.00
- Tmax: Statistics on rate ([0 to 1]) differences : Minimum value: 0.00 Maximum value: 1.97 Mean: 0.02 Standard deviation: 0.13 Median: 0.00 25th percentile (1st quartile): 0.00 75th percentile (3rd quartile): 0.00 99th percentile: 1.00
- TTP: Statistics on rate ([0 to 1]) differences : Minimum value: 0.00 Maximum value: 0.04 Mean: 0.00 Standard deviation: 0.00 Median: 0.00 25th percentile (1st quartile): 0.00 75th percentile (3rd quartile): 0.00 99th percentile: 0.00

So, the differences are fairly small:

CBF :
- Mean: 1%
- Median: 0 %
- 75th percentile: 1%

Difference between 0% and 3%. percentage_difference_CBF_noNorm_CBF

CBV:
- Mean: 0 %
- Median: 0 %
- 75th percentile: 0 %

Difference between 0% and 0.05%. percentage_difference_CBV_noNorm_CBV

MTT:
- Mean: 1 %
- Median: 0 %
- 75th percentile: 1 %

Difference between 0% and 3%. percentage_difference_MTT_noNorm_MTT

T0:
- Mean: 0 %
- Median: 0 %
- 75th percentile: 0 %

Difference between 0% and 0.001%. percentage_difference_T0

Tmax:
- Mean: 2 %
- Median: 0 %
- 75th percentile: 0 %

Difference between 0% and 0.001%. percentage_difference_Tmax

TTP:
- Mean: 0 %
- Median: 0 %
- 75th percentile: 0 %

Difference between 0% and 0.001%. percentage_difference_TTP

servoz commented 4 days ago

During these tests, we tried to use similar input data and identical parameters for the calculations.

If we now run the calculations without changing anything, we get very similar results:

Report obtained with Amigo : report-High-resolution_alej170316_2024-11-08_18h25min14s.pdf

Report obtained with Mia / mia_processes. alej170316_Perf_DSC_Report_2024_11_08_18_27_05_29.pdf

From the point of view of the scripts and the calculations, I think that the pipeline is now correct.

It will be convincing to see the opinion of a radiologist on the results, to definitively validate this pipeline.

I think this ticket can now be closed.

populse / mia_processes

DSC perfusion pipeline validation #87