Add Hybrid Segmentation Loss

[yashika-git]

I was trying to implement UNet++ (issue #896 ). In the UNet++ paper, authors suggest using hybrid segmentation loss:

We then define a hybrid segmentation loss consisting of pixelwise cross-entropy loss and soft dice-coefficient loss for each
semantic scale. The hybrid loss may take advantages of what both loss functions have to offer: smooth gradient and handling
of class imbalance.

Equation 2 in this paper, describes the Mathematical formula to calculate the loss.

Can this loss be implemented in MONAI?

Additional context

[yashika-git]

@Nic-Ma, Shall this loss be added, will it be helpful? Thank you

[Nic-Ma]

Hi @yashika-git ,

I think it would be a nice enhancement for MONAI, welcome your contribution. @wyli Do you have any concern?

Thanks.

[yashika-git]

Thank you, I'll try adding this.

[myron]

is this hybrid loss the same as Dice + CrossEntropy loss? https://github.com/Project-MONAI/MONAI/blob/dev/monai/losses/dice.py#L609

[yashika-git]

@myron, Yes, it seems to be the same. I'll close the issue.