import numpy as np
def _create_grid(radii_mm, spacing, padding_mm):
"""
Helper to create centered physical coordinate grids.
Ensures 0.0 is exactly at the center of the array.
"""
radii_mm = np.array(radii_mm, dtype=np.float64)
spacing = np.array(spacing, dtype=np.float64)
# Field of View limits (radii + padding)
limits = radii_mm + padding_mm
# Create coordinate arrays (centered at 0)
# Using arange ensures steps are exactly 'spacing'
z_mm = np.arange(-limits[0], limits[0], spacing[0])
y_mm = np.arange(-limits[1], limits[1], spacing[1])
x_mm = np.arange(-limits[2], limits[2], spacing[2])
# Create 3D Meshgrid
# indexing='ij' ensures Matrix/Image ordering (Z, Y, X)
Z, Y, X = np.meshgrid(z_mm, y_mm, x_mm, indexing='ij')
return Z, Y, X
[docs]
def cuboid(radii_mm=(30.0, 10.0, 2.5), spacing=(1.0, 1.0, 1.0), padding_mm=5.0):
"""
Creates a binary mask of a solid cuboid.
Args:
radii_mm (tuple): Physical half-lengths (Rx, Ry, Rz) in mm.
Total size will be 2*Rx, 2*Ry, 2*Rz.
spacing (tuple): Voxel spacing (Sz, Sy, Sx) in mm.
padding_mm (float): Padding around the object in mm.
Returns:
image (np.ndarray): Dummy intensity image (value 100 inside).
mask (np.ndarray): Binary mask (uint8).
Example:
>>> from numpyradiomics.dro import cuboid
>>> # Create a 10x10x10mm cube (5mm radii) with 1mm spacing
>>> image, mask = cuboid(radii_mm=(5, 5, 5), spacing=(1.0, 1.0, 1.0))
>>> print(mask.shape)
(20, 20, 20)
"""
Z, Y, X = _create_grid(radii_mm, spacing, padding_mm)
# Cuboid Logic: Intersection of linear bounds
# Using half-open intervals [-R, R) prevents "fencepost" errors where
# centered grids include one too many pixels for even integer dimensions.
mask = (
(Z >= -radii_mm[0]) & (Z < radii_mm[0]) &
(Y >= -radii_mm[1]) & (Y < radii_mm[1]) &
(X >= -radii_mm[2]) & (X < radii_mm[2])
).astype(np.uint8)
image = mask.astype(np.float32) * 100.0
return image, mask
[docs]
def ellipsoid(radii_mm=(30.0, 10.0, 2.5), spacing=(1.0, 1.0, 1.0), padding_mm=5.0):
"""
Creates a binary mask of a solid ellipsoid.
Args:
radii_mm (tuple): Physical radii (Rx, Ry, Rz) in mm.
spacing (tuple): Voxel spacing (Sz, Sy, Sx) in mm.
padding_mm (float): Padding around the object in mm.
Returns:
image (np.ndarray): Dummy intensity image (value 100 inside).
mask (np.ndarray): Binary mask (uint8).
Example:
>>> from numpyradiomics.dro import ellipsoid
>>> # Create an isotropic sphere (radius 10mm) with 0.5mm spacing
>>> image, mask = ellipsoid(radii_mm=(10, 10, 10), spacing=(0.5, 0.5, 0.5))
>>> print(mask.sum()) # Number of voxels in the sphere
33489
"""
Z, Y, X = _create_grid(radii_mm, spacing, padding_mm)
# Ellipsoid Logic: Sum of squared normalized distances <= 1
normalized_dist_sq = (
(Z / radii_mm[0])**2 +
(Y / radii_mm[1])**2 +
(X / radii_mm[2])**2
)
mask = (normalized_dist_sq <= 1.0).astype(np.uint8)
image = mask.astype(np.float32) * 100.0
return image, mask
[docs]
def noisy_ellipsoid(radii_mm=(30.0, 10.0, 2.5), spacing=(1.0, 1.0, 1.0), padding_mm=5.0, intensity_range=(0, 100)):
"""
Creates a binary mask of an ellipsoid filled with random noise.
Useful for testing texture features.
Args:
radii_mm (tuple): Physical radii (Rx, Ry, Rz) in mm.
spacing (tuple): Voxel spacing (Sz, Sy, Sx) in mm.
padding_mm (float): Padding around the object in mm.
intensity_range (tuple): (min, max) intensity values for noise.
Returns:
image (np.ndarray): Noisy image (float32).
mask (np.ndarray): Binary mask (uint8).
Example:
>>> from numpyradiomics.dro import noisy_ellipsoid
>>> import numpy as np
>>> # Create a noisy ellipsoid for texture analysis
>>> img, mask = noisy_ellipsoid(radii_mm=(20, 10, 5))
>>> print(f"Mean Intensity in ROI: {np.mean(img[mask==1]):.2f}")
Mean Intensity in ROI: 49.87
"""
Z, Y, X = _create_grid(radii_mm, spacing, padding_mm)
# Ellipsoid Logic
normalized_dist_sq = (
(Z / radii_mm[0])**2 +
(Y / radii_mm[1])**2 +
(X / radii_mm[2])**2
)
mask = (normalized_dist_sq <= 1.0).astype(np.uint8)
# Generate Noise
np.random.seed(42)
image = np.random.uniform(intensity_range[0], intensity_range[1], mask.shape).astype(np.float32)
# Apply Mask (Background = 0)
image[mask == 0] = 0
return image, mask
