Error when running brats segmentation model (v 0.3.8) from monaibundle

[tdakinosho]

Hello community,

I am currently having troubles running the brats model for brain segmentation on 3D slicer. I have a custom dataset(originally DICOM images but converted to NIFTI using SimpleiTK library) that i am initially trying to run the model on before i commence training but i am stuck with this. Everything works fine when i tried using deepedit from the radiology app but since my dataset is of brain scans, i would rather have the brats model work fine. I have tried going through the suggestions on this discussion thread (#1051) but i could not find a solution to my error. Please find the error i am experiencing below and what i have tried so far.

Error:
RuntimeError: Error(s) in loading state_dict for SegResNet:
size mismatch for convInit.conv.weight: copying a param with shape torch.Size([16, 4, 3, 3, 3]) from checkpoint, the shape in current model is torch.Size([16, 1, 3, 3, 3]).

What i have tried so far:
Updated the inference.json in monaibundle\brats_mri_segmentation_v0.3.8\configs with 'ensure_channel_first' arg
Changed the "in_channels" value to 1 from 4 in brats_mri_segmentation_v0.3.8/configs/train.json ,inference.json and metadata.json.
I have tried downloading and using the 'Task01_BrainTumour' dataset but i got the same error

Please any help or suggestions would be appreciated

[diazandr3s]

Hi @tdakinosho,

Sorry I completely missed your question.

Can I ask how many modalities are you using for this? All four as in the BRATS dataset? Have you tried using the MONAI Auto3DSeg module in Slicer?

Just download the latest Slicer and then install Auto3DSeg via the Extension Manager

Please let us know,