Processing Operations

The medical_image.process module provides stateless image processing operations. All methods are static and decorated with @requires_loaded — they raise an error if the input image has not been loaded.

General Pattern

Every processing method follows the same signature:

SomeProcessor.operation(image, output, <params>, device=None)

• image — input Image (read-only)

• output — output Image (result written here)

• device — optional; inferred from image if omitted

Filters

Spatial and frequency domain filters:

from medical_image import Filters

output = image.clone()

# Gaussian blur
Filters.gaussian_filter(image, output, sigma=2.0)

# Median filter (denoising)
Filters.median_filter(image, output, size=5)

# Difference of Gaussians (band-pass)
Filters.difference_of_gaussian(image, output, low_sigma=1.7, high_sigma=2.0)

# Laplacian of Gaussian (edge detection)
Filters.laplacian_of_gaussian(image, output, sigma=1.5)

# Gamma correction
Filters.gamma_correction(image, output, gamma=0.5)

# Contrast adjustment
Filters.contrast_adjust(image, output, contrast=1.5, brightness=0.1)

Thresholding

Global and adaptive thresholding:

from medical_image import Threshold

output = image.clone()

# Otsu (global, automatic threshold)
Threshold.otsu_threshold(image, output)

# Sauvola (adaptive local threshold)
Threshold.sauvola_threshold(image, output, window_size=15, k=0.2)

Morphological Operations

Binary and grayscale morphology:

from medical_image import MorphologyOperations

output = image.clone()

# Closing (fills small gaps)
MorphologyOperations.morphology_closing(image, output, kernel_size=7)

# Erosion / Dilation (disk structuring element)
MorphologyOperations.erosion(image, output, radius=2)
MorphologyOperations.dilation(image, output, radius=3)

# White top-hat (isolates bright structures)
MorphologyOperations.white_top_hat(image, output, radius=4)

# Region fill (fill holes in binary mask)
MorphologyOperations.region_fill(image, output)

Frequency Domain

FFT-based operations:

from medical_image import FrequencyOperations

output = image.clone()
FrequencyOperations.fft(image, output)
# output.pixel_data now holds the magnitude spectrum

FrequencyOperations.inverse_fft(output, reconstructed)

Metrics

Information-theoretic metrics:

from medical_image import Metrics

# Shannon entropy
h = Metrics.entropy(image)

# Joint entropy of two images
h_joint = Metrics.joint_entropy(image1, image2)

# Mutual information
mi = Metrics.mutual_information(image1, image2)

# Variance (global or local)
var = Metrics.variance(image)

Mammography-Specific

Specialized operations for mammogram preprocessing:

from medical_image import MammographyPreprocessing

# Extract breast region mask
mask = MammographyPreprocessing.breast_mask(image)

# Apply DICOM windowing
windowed = MammographyPreprocessing.dicom_window(
    image, window_center=2000, window_width=3000
)

# Normalize bit depth (12-bit DICOM -> [0, 255])
normalized = MammographyPreprocessing.normalize_bit_depth(
    image, bits_stored=12
)