PatchDataset/GridPatchDataset or SlidingWindowInferer for patch-based train with large volume

[Littlehhh]

Description

Hi
I found that there are several ways in MONAI to implement patch-based(grid patch) training for large medical volume.

1. PatchDataset/GridPatchDataset with SimpleInferer
2. Dataset with SlidingWindowInferer
3. Dataset with RandCropByPosNegLabel/(or extra transfrom)

So which way should be better? And these ways have different loss(single patch/full volume/several patch) backpropagations.
I am confused about this and I think maybe the first way is a more reasonable way for patch-based system.

In addition, from the perspective of framework design, which way is more appropriate, and what are the uses of other ways for patch-based train?

Thanks.

[Nic-Ma]

Hi @Littlehhh ,

Thanks for your interest here.

1. RandCropByPosNegLabel is a transform and usually used to generate random patches during training, the following transforms will apply to every patch separately.
2. SlidingWindowInfer is used to split big image into patches for the network forward(), so it doesn't contain random logic and the input data is already handled by transforms, usually used for validation.
3. PatchDataset can do global shuffling on the patches.

You can start from this example:
https://github.com/Project-MONAI/tutorials/blob/master/3d_segmentation/spleen_segmentation_3d.ipynb

Thanks.

[Littlehhh]

Hi @Nic-Ma
Thanks for your reply.

Both PatchDataset and SlidingWindowInferer are for large volume patch-based process.
Is it possible to integrated these two functions together ？
PatchDataset has already prepared the patch so SlidingWindowInferer is a little bit redundant and it introduces one more module to the MONAI. Is this module necessary for MONAI.

Thanks in advance.

[Nic-Ma]

Hi @Littlehhh ,

SlidingWindow is a very common inference method, especially for medical images, I think it's necessary to provide this feature.

Thanks.

[Littlehhh]

Hi @Nic-Ma
Thanks for your reply.
Very happy to discuss this with you.
I agree that SlidingWindow is a important function for medical images.
Is there any way to integrate these two component together？
Just like check if PatchDataset than use slidingwindow.
but PatchDataset has already cropped the patch, and slidingwindow do the same crop.
I think maybe there is a suitable way for Patch-based train and Inference.

Thanks in advance.

[Nic-Ma]

Hi @wyli ,

Could you please help share more use cases about PatchDataset?

Thanks in advance.