search

EIDOSLAB / Neural-Network-derived-perfusion-maps

1 stars 2 forks source link

best_model_NLR_CBV.pt #2

Open hetao255 opened 1 year ago

hetao255 commented 1 year ago

Hello, sorry to disturb you again. I tried to train the model on this code, but the results seemed unsatisfactory. Could you please provide one of your trained pt files for direct testing? The graph below is the one of the outputs of my training process. MOL_087_NLR_CBV_val_epoch_30_14

brechmos commented 1 year ago

We found a very similar result here too. @hetao255 were you able to find a fix for this?

I am wondering if there is a skull stripping (or brain extraction) step that needs to happen on the input CTP data that is not part of this repo (?). All the ground truth masks (e.g., on the left above) appear to be skull stripped but the predicted on the right (and similar to what we see) implies the network is not learning a skull stripping (or brain extraction).