Finetuning for 3 class segmentation

[salohiddin22]

Hi Reza. Thanks for the detailed explanation in your blog and the code.

I was trying to use your code for 3 class segmentation on the whole image. For the binary it worked fine.

I mainly changed the following parts.

• The bounding box to the image size.
• ScaleIntensityRanged(keys=['label'], a_min=53, a_max=226, b_min=0.0, b_max=2.0, clip=True),
• seg_loss = monai.losses.DiceCELoss(to_onehot_y=True, softmax=True, squared_pred=True, reduction='mean')".

But I cannot get any results it's just giving black image as prediction.

Can you suggest anything to fit it for multi-class segmentation?

Many thanks!