feat: add N4BiasFieldCorrection transform class

src/imgtools/transforms/__init__.py

@@ -11,6 +11,7 @@
 from .intensity_transforms import (
     ClipIntensity,
     IntensityTransform,
+    N4BiasFieldCorrection,
     WindowIntensity,
 )
 from .lambda_transforms import ImageFunction, SimpleITKFilter
@@ -42,6 +43,7 @@
     "IntensityTransform",
     "ClipIntensity",
     "WindowIntensity",
+    "N4BiasFieldCorrection",
     # spatial transform
     "SpatialTransform",
     "Resample",

src/imgtools/transforms/functional.py

@@ -1,6 +1,8 @@
 import numpy as np
 import SimpleITK as sitk
 
+from .lambda_transforms import SimpleITKFilter
+
 INTERPOLATORS = {
     "linear": sitk.sitkLinear,
     "nearest": sitk.sitkNearestNeighbor,
@@ -13,6 +15,7 @@
     "zoom",
     "rotate",
     "crop",
+    "bias_correction",
     "clip_intensity",
     "window_intensity",
 ]
@@ -410,3 +413,21 @@ def window_intensity(
     lower = level - window / 2
     upper = level + window / 2
     return clip_intensity(image, lower, upper)
+
+
+def bias_correction(image: sitk.Image) -> sitk.Image:
+    """Apply N4 bias field correction to reduce smooth intensity inhomogeneities (bias fields) commonly found in
+    MR imaging. This transform corrects voxel intensities while preserving image geometry (spacing, orientation, and dimensions).
+
+    Parameters
+    ----------
+    image : sitk.Image
+        The MR image to apply bias field correction to.
+
+    Returns
+    -------
+    sitk.Image
+        The bias corrected image
+    """
+    bias_filter = SimpleITKFilter(sitk.N4BiasFieldCorrectionImageFilter())
+    return bias_filter(image)

src/imgtools/transforms/intensity_transforms.py

@@ -3,15 +3,13 @@
 from SimpleITK import Image
 
 from .base_transform import BaseTransform
-from .functional import (
-    clip_intensity,
-    window_intensity,
-)
+from .functional import bias_correction, clip_intensity, window_intensity
 
 __all__ = [
     "IntensityTransform",
     "ClipIntensity",
     "WindowIntensity",
+    "N4BiasFieldCorrection",
 ]
 
 
@@ -150,3 +148,42 @@ def __call__(self, image: Image) -> Image:
         """
 
         return window_intensity(image, self.window, self.level)
+
+
+@dataclass
+class N4BiasFieldCorrection(IntensityTransform):
+    """N4BiasFieldCorrection operation class.
+
+    A callable class that applies N4 bias field correction to reduce
+    smooth intensity inhomogeneities (bias fields) commonly found in
+    MR imaging. This transform corrects voxel intensities while
+    preserving image geometry (spacing, orientation, and dimensions).
+
+    The correction uses SimpleITK's N4BiasFieldCorrectionImageFilter,
+    which implements the N4 algorithm (Tustison et al., 2010).
+
+    Notes
+    -----
+    - This transform is computationally intensive and may take several
+      seconds to minutes depending on image size.
+    - Best suited for MR images; application to other modalities is
+      typically not meaningful.
+    - No rotation, translation, or scaling is applied.
+    """
+
+    # def __post_init__(self) -> None:
+    #     self.filter = SimpleITKFilter(N4BiasFieldCorrectionImageFilter())
+
+    def __call__(self, image: Image) -> Image:
+        """Apply N4 bias-field correction to an image.
+
+        Parameters
+        ----------
+        image : Image
+            The input MR image to apply bias-field correction.
+        Returns
+        -------
+        Image
+            The MR image after applying N4 bias field correction.
+        """
+        return bias_correction(image)

src/imgtools/transforms/transformer.py

@@ -11,6 +11,7 @@
     IntensityTransform,
     SpatialTransform,
 )
+from imgtools.transforms.intensity_transforms import N4BiasFieldCorrection
 
 # Define TypeVars for the different image types
 T_MedImage = TypeVar("T_MedImage", bound=MedImage)
@@ -71,7 +72,17 @@ def _apply_transforms(
                 elif isinstance(
                     transform, (IntensityTransform, SpatialTransform)
                 ):
-                    transformed_image = transform(transformed_image)
+                    # Apply N4BiasFieldCorrection only for MR images
+                    if isinstance(transform, N4BiasFieldCorrection):
+                        if (
+                            transformed_image.metadata.get(
+                                "Modality", "Unknown"
+                            )
+                            == "MR"
+                        ):
+                            transformed_image = transform(transformed_image)
+                    else:
+                        transformed_image = transform(transformed_image)
                 else:
                     msg = f"Invalid transform type: {type(transform)}"
                     raise ValueError(msg)