Results on nnU-Net trained over 45 subjects mixing healthy and mild compression subjects

Question

Results on nnU-Net trained over 45 subjects mixing healthy and mild compression subjects

abelsalm opened this issue 4 months ago · 5 comments

Here is a review on the last training I did on ann nnU-Net for spinal canal segmentation.
I added more data using the previous training, 10th issue, trained the model on 45 subjetcs using the results of the inference, mixing data from spinge-generic, dcm-oklahoma, dcm-brno, sci-paris, and dcm zurich.

I trained the model for 400 epochs, with the randomly generated fold done by nnUNettrainer function.

So there were 5 subjetcs for the testing part, one from each dataset, and 40 for the training, I used the first fold, in this json file :
splits_final.json

Here are the details of the training :

May be I could updload the qc report, anyway I already did more active learning and trained a new model on more data again, see in the next issue.

Answer 1 · 2024-09-09T18:52:32.000Z

@jcohenadad @sandrinebedard @valosekj
May be you think I should add things on this issue ?
Anyway I'm opening an other one for the training I finished last week on 97 subjects.

Answer 2 · 2024-09-10T20:49:55.000Z

You can also maybe specify that spinge-generic, the data is derivatives/data_preprocessed, and the versions of each dataset

Answer 3 · 2024-09-11T18:15:52.000Z

Here is the QC for this session : https://drive.google.com/file/d/1-LQyUpK4OCnyP2akaeJCiM7Dyeq1acwO/view?usp=drive_link

Answer 4 · 2024-09-11T19:23:43.000Z

cool! some issues, notably:

sub-3194B_ses-3194B_T2w_000_0000.nii.gz
sub-3249B_ses-3249B_T2w_000_0000.nii.gz

but overall it looks ok

Answer 5 · 2024-09-11T19:32:51.000Z

cool! some issues, notably:

sub-3194B_ses-3194B_T2w_000_0000.nii.gz
You mean on the superior part ? cause I don't see anything else ?