Segmentation and Classification Task

[Meddebma]

Hi,
I successfully segmented spleens in my dataset using the following code:
https://github.com/Meddebma/pyradiomics/blob/master/3D_Segmentation_eval.ipynb

In a second step I tried to classify the segmentation results into two labels in this code:
https://github.com/Meddebma/pyradiomics/blob/master/3D_Classification_Dict.ipynb

unfortunately I've got an error concerning the segmentations which have for example the shape of (512, 512, 50, 2), how can I get rid of the last dimension(2) or how should I manipulate the data to do the classification task?

Thanks many times

[Nic-Ma]

Hi @Meddebma ,

Thanks for your interest here with MONAI APIs.
Sorry I don't quite understand your question. Do you mean to move your last dim(channel) to be first dim?
You can use EnsureChannelFirst transform or AsChannelFirst transform.

Thanks.

[rijobro]

Sure. I suspect that the following should simply work:

saver.save_batch(val_outputs[:, 0, ...], val_data["image_meta_dict"])

i.e., all batches, 0th channel, all spatial dimensions.

Let me know if that doesn't work,

[Meddebma]

@rijobro,
I tried it, it took very long time writing the file but then the system crashed
https://github.com/Meddebma/pyradiomics/blob/master/3D_Segmentation_eval.ipynb

[rijobro]

can't see any errors

[Meddebma]

okey, and if I try to discard the background in the classification task,
where should I put
images = images[:, 0, ...]?

https://github.com/Meddebma/pyradiomics/blob/master/3D_Classification.ipynb

[rijobro]

if you want to remove the channel when reading them in, you'll need to do something like:

t = Compose([
    LoadImage
    AsChannelFirst
    Lambda(lambda x: x[0])

Notice how during the read step, we use [0] rather than [:, 0, ...] because during the post-process we have the batch dimension, but during the reading, we don't.