A question regarding the testing on BTCV

[wwzjer]

Hi author,

I was a little bit confusing about the test configuration for BTCV. I found in your code when you load the test data, you used the following augmentation/transforms:

    val_transform = transforms.Compose(
        [
            transforms.LoadImaged(keys=["image", "label"]),
            transforms.EnsureChannelFirstd(keys=["image", "label"]),
            transforms.Orientationd(keys=["image", "label"], axcodes="RAS"),
            transforms.Spacingd(
                keys=["image", "label"], pixdim=(args.space_x, args.space_y, args.space_z), mode=("bilinear", "nearest")
            ),
            transforms.ScaleIntensityRanged(
                keys=["image"], a_min=args.a_min, a_max=args.a_max, b_min=args.b_min, b_max=args.b_max, clip=True
            ),
            transforms.CropForegroundd(keys=["image", "label"], source_key="image"),
            # transforms.ToTensord(keys=["image", "label"]),
        ]
    )

When I check the code of SwinUNETR, their transform seems different:

    test_transform = transforms.Compose(
        [
            transforms.LoadImaged(keys=["image", "label"]),
            transforms.AddChanneld(keys=["image", "label"]),
            # transforms.Orientationd(keys=["image"], axcodes="RAS"),
            transforms.Spacingd(keys="image", pixdim=(args.space_x, args.space_y, args.space_z), mode="bilinear"),
            transforms.ScaleIntensityRanged(
                keys=["image"], a_min=args.a_min, a_max=args.a_max, b_min=args.b_min, b_max=args.b_max, clip=True
            ),
            transforms.ToTensord(keys=["image", "label"]),
        ]
    )

I found it may induce an unignorable performance difference. I was wondering if "transforms.Orientationd(keys=["image", "label"], axcodes="RAS")" or "transforms.CropForegroundd(keys=["image", "label"], source_key="image")" can be used for test data as the label information should be unknown. I would appreciate it if you could help me with this. Thanks!

[Luffy03]

Hi, many thanks for your attention to our work. It is a pity that the BTCV test leaderboard is no longer updating now. So in this repo we only provide the codes for validation. The test_transform is actually "val_transform". If you want to conduct test on some other datasets, you only need to remove the key "label".

[wwzjer]

Thanks for your reply. I was wondering in your paper, are you using val_transform (same as you did) or test_transform for the comparison methods. Thanks.

[Luffy03]

Hi, sorry, I am a little confused with your question. Val_transform is used for val set and test transform is used for test sets (e.g., AMOS with test learderboard). Does this response align with what you were looking for?

[Luffy03]

Dear researcher, our work is now available at Large-Scale-Medical, if you are still interested in this topic. Thank you very much for your attention to our work, it does encourage me a lot!