Lung Segmentation output very different from label

[jonathanquantumalpha]

Hi,

I'm doing lung segmentation from the medical decathlon data for lung. The code works fine without any errors, however the results I'm getting is very different from the labels (see attached image -- basically the output is highlighting the whole lung as tumorous). Here are the current transforms that I have:

        [
            LoadImaged(keys=["vol", "seg"]),
            AddChanneld(keys=["vol", "seg"]),
            Orientationd(keys=["vol", "seg"], axcodes="RAS"),
            ScaleIntensityRanged(keys=["vol"], a_min=a_min, a_max=a_max, b_min=0.0, b_max=1.0, clip=True),
            CropForegroundd(keys=["vol", "seg"], source_key="vol"),
            Resized(keys=["vol", "seg"], spatial_size=spatial_size),
            ToTensord(keys=["vol", "seg"]),

        ]
    )

    test_transforms = Compose(
        [
            LoadImaged(keys=["vol", "seg"]),
            AddChanneld(keys=["vol", "seg"]),
            Orientationd(keys=["vol", "seg"], axcodes="RAS"),
            ScaleIntensityRanged(keys=["vol"], a_min=a_min, a_max=a_max,b_min=0.0, b_max=1.0, clip=True),
            CropForegroundd(keys=['vol', 'seg'], source_key='vol'),
            Resized(keys=["vol", "seg"], spatial_size=spatial_size),
            ToTensord(keys=["vol", "seg"]),


        ]```

Would appreciate any ideas on how I can improve this. Thanks!
![download (1)](https://github.com/Project-MONAI/MONAI/assets/135941084/d9fb69ce-e35f-48e6-992f-3e00f90d999b)

[KumoLiu]

Hi @jonathanquantumalpha, there are many possible reasons.
First I would suggest you follow the tutorial.
After pre-processing, I also suggest you check your input whether makes sense.
Also, you could modify your hyperparameters such as learning rate, optimizer, et.al.
Thanks!