Add more transforms

spleenseg/transforms/transforms.py

@@ -28,20 +28,58 @@
 from monai.transforms.croppad.dictionary import CropForegroundd, RandCropByPosNegLabeld
 from monai.transforms.spatial.dictionary import Orientationd, RandAffined, Spacingd
 from monai.transforms.intensity.dictionary import ScaleIntensityRanged
-from typing import Any, Optional, Callable, Hashable, Mapping, Dict, Union
+from monai.transforms.io.array import LoadImage
+from monai.config.type_definitions import PathLike
+from typing import Any, Callable, Hashable, Mapping, Sequence
+from monai.data.meta_tensor import MetaTensor
 import numpy as np
+from numpy import ndarray
 from pathlib import Path
 from spleenseg.plotting import plotting
 
 
-def f_LoadImaged() -> Callable[[dict[str, Any]], dict[str, Any]]:
-    return LoadImaged(keys=["image", "label"])
+def f_LoadImaged(
+    keys: list[str] = ["image", "label"],
+) -> Callable[[dict[str, Any]], dict[str, Any]]:
+    return LoadImaged(keys=keys)
+
+
+def f_LoadImage() -> Callable[
+    [PathLike | Sequence[PathLike]],
+    torch.Tensor
+    | Any
+    | MetaTensor
+    | tuple[
+        torch.Tensor | Any | MetaTensor,
+        dict[Any, Any]
+        | Any
+        | ndarray[Any, Any]
+        | tuple[Any, ...]
+        | list[Any]
+        | bool
+        | str
+        | float
+        | int
+        | None,
+    ],
+]:
+    return LoadImage()
+
+
+def f_SaveImaged(outputDir: Path) -> Callable[[dict[str, Any]], dict[str, Any]]:
+    return SaveImaged(
+        keys="pred",
+        meta_keys="pred_meta_dict",
+        output_dir=str(outputDir),
+        output_postfix="seg",
+        resample=False,
+    )
 
 
-def f_EnsureChannelFirstd() -> (
-    Callable[[Mapping[Hashable, torch.Tensor]], Mapping[Hashable, torch.Tensor]]
-):
-    return EnsureChannelFirstd(keys=["image", "label"])
+def f_EnsureChannelFirstd(
+    keys: list[str] = ["image", "label"],
+) -> Callable[[Mapping[Hashable, torch.Tensor]], Mapping[Hashable, torch.Tensor]]:
+    return EnsureChannelFirstd(keys=keys)
 
 
 def f_ScaleIntensityRanged() -> (
@@ -52,25 +90,26 @@ def f_ScaleIntensityRanged() -> (
     )
 
 
-def f_CropForegroundd() -> (
-    Callable[[Mapping[Hashable, torch.Tensor]], Mapping[Hashable, torch.Tensor]]
-):
-    return CropForegroundd(keys=["image", "label"], source_key="image")
+def f_CropForegroundd(
+    keys: list[str] = ["image", "label"],
+) -> Callable[[Mapping[Hashable, torch.Tensor]], Mapping[Hashable, torch.Tensor]]:
+    return CropForegroundd(keys=keys, source_key="image", allow_smaller=True)
 
 
-def f_Orientationd() -> (
-    Callable[[Mapping[Hashable, torch.Tensor]], Mapping[Hashable, torch.Tensor]]
-):
-    return Orientationd(keys=["image", "label"], axcodes="RAS")
+def f_Orientationd(
+    keys: list[str] = ["image", "label"],
+) -> Callable[[Mapping[Hashable, torch.Tensor]], Mapping[Hashable, torch.Tensor]]:
+    return Orientationd(keys=keys, axcodes="RAS")
 
 
-def f_Spaceingd() -> (
-    Callable[[Mapping[Hashable, torch.Tensor]], Mapping[Hashable, torch.Tensor]]
-):
+def f_Spaceingd(
+    keys: list[str] = ["image", "label"],
+    mode: tuple[str, ...] = ("bilinear", "nearest"),
+) -> Callable[[Mapping[Hashable, torch.Tensor]], Mapping[Hashable, torch.Tensor]]:
     return Spacingd(
-        keys=["image", "label"],
+        keys=keys,
         pixdim=(1.5, 1.5, 2.0),
-        mode=("bilinear", "nearest"),
+        mode=mode,
     )
 
 
@@ -162,6 +201,30 @@ def trainingAndValidation_transformsSetup() -> tuple[Compose, Compose]:
     return trainingTransforms, validationTransforms
 
 
+def validation_transformsOnOriginal() -> Compose:
+    transforms: list = [
+        f_LoadImaged(),
+        f_EnsureChannelFirstd(),
+        f_Orientationd(["image"]),
+        f_Spaceingd(["image"], tuple(["bilinear"])),
+        f_ScaleIntensityRanged(),
+        f_CropForegroundd(["image"]),
+    ]
+    return transforms_build(transforms)
+
+
+def inferenceUse_transforms() -> Compose:
+    transforms: list = [
+        f_LoadImaged(["image"]),
+        f_EnsureChannelFirstd(["image"]),
+        f_Orientationd(["image"]),
+        f_Spaceingd(["image"], tuple(["bilinear"])),
+        f_ScaleIntensityRanged(),
+        f_CropForegroundd(["image"]),
+    ]
+    return transforms_build(transforms)
+
+
 def transforms_check(
     outputdir: Path, files: list[dict[str, str]], transforms: Compose
 ) -> bool:
@@ -171,6 +234,9 @@ def transforms_check(
     if not check_data:
         return False
     image, label = (check_data["image"][0][0], check_data["label"][0][0])
-    print(f"image shape: {image.shape}, label shape: {label.shape}")
+    print("")
+    print("Checking transforms... :")
+    print(f"sample image shape: {image.shape}")
+    print(f"sample label shape: {label.shape}")
     plotting.plot_imageAndLabel(image, label, outputdir / "exemplar_image_label.jpg")
     return True