Lung tumor segmentation

[Meddebma]

Hi,

I am trying to implement the 3d-segmentation pipeline with the Lung_Dataset of MSD, and I was wondering if there are transforms specific to Lung tumor segmentation, I combined transforms of the registration and the challenge tutorials
here are the transforms I used:

train_transforms = Compose(
[
LoadImaged(
keys=["image", "label"]
),
AddChanneld(
keys=["image", "label"]
),
Orientationd(keys=["image", "label"], axcodes="LPS"),
Spacingd(keys=["image", "label"], pixdim=(1.25, 1.25, 5.0), mode=("bilinear", "nearest")[: len(["image", "label"])]),
ScaleIntensityRanged(
keys=["image", "label"],
a_min=-285, a_max=3770, b_min=0.0, b_max=1.0, clip=True,
),
SpatialPadd(keys=["image", "label"], spatial_size=(192, 192, -1), mode="reflect"),
RandAffined(
keys=["image", "label"],
mode=('bilinear', 'nearest'),
prob=1.0, spatial_size=(192, 192, 208),
rotate_range=(0, 0, np.pi / 15), scale_range=(0.1, 0.1, 0.1)
),
RandCropByPosNegLabeld(keys=["image", "label"], label_key=["label"], spatial_size=(192, 192, 16), num_samples=3),
Resized(
keys=["image", "label"],
mode=('trilinear', 'nearest'),
align_corners=(True, None),
spatial_size=(96, 96, 104)
),
EnsureTyped(
keys=["image", "label"]
),
])

When I was loading the data, there was an error implementing RandCropByPosNegLabel, does it have something to do with the spacial size? Which transforms should I keep for this task? thank you very much!

[Nic-Ma]

Hi @wyli @ericspod @dongyang0122 ,

Could you please help share some training experience with Lung data?

Thanks in advance.

[dongyang0122]

Hi @Meddebma , you may use the following tutorial for MSD tasks. That would a good starting point.

https://github.com/Project-MONAI/tutorials/tree/master/modules/dynunet_pipeline