Data preprocess

[b3r-prog]

Hi @himashi92

Thanks for this awesome work and repo. I am trying to replicate this on the Liver data in the decathlon. When I run the the train command for the base training CUDA_VISIBLE_DEVICES=0 nohup vtunet_train 3d_fullres vtunetTrainerV2_vtunet_tumor_base 3 0 &> base.out &

I get the following error below KeyError: 'BRATS_001'():
`stage: 0
{'batch_size': 2, 'num_pool_per_axis': [5, 5, 5], 'patch_size': array([128, 128, 128]), 'median_patient_size_in_voxels': array([195, 207, 207]), 'current_spacing': array([2.473119 , 1.89831205, 1.89831205]), 'original_spacing': array([1. , 0.76757812, 0.76757812]), 'do_dummy_2D_data_aug': False, 'pool_op_kernel_sizes': [[2, 2, 2], [2, 2, 2], [2, 2, 2], [2, 2, 2], [2, 2, 2]], 'conv_kernel_sizes': [[3, 3, 3], [3, 3, 3], [3, 3, 3], [3, 3, 3], [3, 3, 3], [3, 3, 3]]}

stage: 1
{'batch_size': 2, 'num_pool_per_axis': [5, 5, 5], 'patch_size': array([128, 128, 128]), 'median_patient_size_in_voxels': array([482, 512, 512]), 'current_spacing': array([1. , 0.76757812, 0.76757812]), 'original_spacing': array([1. , 0.76757812, 0.76757812]), 'do_dummy_2D_data_aug': False, 'pool_op_kernel_sizes': [[2, 2, 2], [2, 2, 2], [2, 2, 2], [2, 2, 2], [2, 2, 2]], 'conv_kernel_sizes': [[3, 3, 3], [3, 3, 3], [3, 3, 3], [3, 3, 3], [3, 3, 3], [3, 3, 3]]}

I am using stage 1 from these plans
I am using batch dice + CE loss

I am using data from this folder: /home/VT-UNet/VTUNet/DATASET/vtunet_preprocessed/Task003_Liver/vtunetData_plans_v2.1
###############################################
loading dataset
loading all case properties
2022-08-16 04:36:37.549490: Creating new 5-fold cross-validation split...
2022-08-16 04:36:37.550782: Desired fold for training: 0
2022-08-16 04:36:37.550879: This split has 387 training and 73 validation cases.
Traceback (most recent call last):
File "/home/VT-UNet/transegvenv/bin/vtunet_train", line 33, in
sys.exit(load_entry_point('vtunet', 'console_scripts', 'vtunet_train')())
File "/home/VT-UNet/VTUNet/vtunet/run/run_training.py", line 134, in main
trainer.initialize(not validation_only)
File "/home/VT-UNet/VTUNet/vtunet/training/network_training/vtunetTrainerV2_vtunet_liver_base.py", line 90, in initialize
self.dl_tr, self.dl_val = self.get_basic_generators()
File "/home/VT-UNet/VTUNet/vtunet/training/network_training/vtunetTrainer.py", line 401, in get_basic_generators
self.do_split()
File "/home/VT-UNet/VTUNet/vtunet/training/network_training/vtunetTrainerV2_vtunet_liver_base.py", line 410, in do_split
self.dataset_tr[i] = self.dataset[i]
KeyError: 'BRATS_001'`

Any idea how to fix this, also but separately I get an error when I try to do the vtunet_train_3d?

[himashi92]

Thank you. Inside your vtunetTrainerV2_vtunet_liver_base.py script, modify the file name prefix in the code line:

splits[self.fold]['train'] = np.array(['LiverXX', ... ])

[b3r-prog]

Thanks for the reply. Just to clarify I should change all the prefix for the data splits?

i.e. change from this splits[self.fold]['train'] = np.array(['BRATS_001', 'BRATS_002', 'BRATS_003', 'BRATS_004' ...

to this

splits[self.fold]['train'] = np.array(['liver_001', 'liver_002', 'liver_003', 'liver_004' .... same convention in the preprocessed files.

[b3r-prog]

fyi @himashi92, I made the change but I'm still getting this error

RuntimeError: Could not find trainer class in vtunet.training.network_training. I'm updating the class function in the vtunetTrainerV2_vtunet_liver_base.py file to vtunetTrainerV2_vtunet_liver_base from vtunetTrainerV2_vtunet_tumor_base. I guess I have to change the classes as well

[b3r-prog]

I was able to resolve the above errors but now getting this KeyError: 'liver_131'. When I check the preprocessed files, it seems to stop at liver_130. Is there a way to change this to ensure that all the data is preprocessed? Right now, the file vtunetTrainerV2_vtunet_liver_base.py shows that the data should go all the way to liver_484 but the preprocessed files stop at liver_130.

I understand a little now. The vtunetTrainerV2_vtunet_liver_base.py file generates based on both train and test while the output data splits is correctly only on the train segment. I'll remove the unneeded entries from the vtunetTrainerV2_vtunet_liver_base.py and see if it works.

[b3r-prog]

Hi @himashi92, resolved above but getting this error RuntimeError: Given groups=1, weight of size [72, 4, 4, 4, 4], expected input[2, 1, 128, 128, 128] to have 4 channels, but got 1 channels instead now. I guess this is related to the number of channels difference between the liver and tumor images. Is there a specific place to change this? Your previous comment suggested changing here VT-UNet/vtunet/vision_transformer.py but the file is no longer available. I can also help with a write up on the readme on adapting to a different dataset

[himashi92]

create a script like vtunet_tumor.py for liver dataset and change in_channels=1 https://github.com/himashi92/VT-UNet/blob/main/VTUNet/vtunet/network_architecture/vtunet_tumor.py#L780

[b3r-prog]

Thanks, works now.