Compute CSA prediction error

[sandrinebedard]

Description

In the analysis of CSA, we want to compute the CSA error between the GT and the prediction for every image of the test set:

ABS(CSA_pred- CSA_GT)

TODO

• [x] Set the same Y lim between methods
• [x] Validate to take GT csa_GT_soft_bin
• [x] For violin plot to compare all methods error, append all contrasts in the same column instead of mean

[sandrinebedard]

Here are some preliminary results:

Error for csa_monai_csaDiceL_160x224x96:

Error for ivado_avg_bin_no_crop:

Error for csa_nnunet_soft_avg_all_no_crop:

For all:

**note that the mean per subject is done, which may not relfect the max values --> in TODOS

Note that for now it is not Binarized CSA soft GT that is used, this may change the results...

[jcohenadad]

great results! I’m glad we beat nnUNET and ivadomed

[naga-karthik]

Thanks for creating these plots @sandrinebedard ! Just to clarify:

In the analysis of CSA, we want to compute the CSA error between the GT and the prediction for every image of the test set: ABS(CSA_pred- CSA_GT)

This is the CSA of the entire label right? Not just the CSA of C2-C3 levels?

[sandrinebedard]

Thanks for creating these plots @sandrinebedard ! Just to clarify:

In the analysis of CSA, we want to compute the CSA error between the GT and the prediction for every image of the test set: ABS(CSA_pred- CSA_GT)

This is the CSA of the entire label right? Not just the CSA of C2-C3 levels?

This is at C2-C3, you are right is would be good to compte the error on the entire mask, or slicewise.

[naga-karthik]

Closing as all tasks are done and we have the latest plots