Skip to content

Texture features

The TextureFeatures class provides an efficient mechanism for calculating second-order texture features from a given 3D magnetic resonance image (MRI).

Overview

This class utilizes skimage for calculating the gray level co-occurrence matrix (GLCM) and its corresponding texture features such as contrast, homogeneity, and energy. The texture features extracted from each 2D plane of a 3D MRI sequence give insights into the structural patterns within the image.

By encapsulating these operations in a class, the user can easily compute several texture features with minimal effort. It also supports an option to remove empty planes, improving accuracy when working with brain MRI scans.

The following texture features are available:

  • Contrast: A measure of the intensity contrast between a pixel and its neighbor over the whole image.
  • Dissimilarity: Measures the local intensity variations.
  • Homogeneity: Measures the closeness of the distribution of elements in the GLCM to the GLCM diagonal.
  • ASM (Angular Second Moment): A measure of the texture uniformity.
  • Energy: The square root of ASM, indicating the texture’s level of orderliness.
  • Correlation: A measure of how correlated a pixel is to its neighbor across the whole image.

Methods

__init__

Constructs all the necessary attributes for the TextureFeatures object.

Parameters

  • sequence (np.ndarray): A 3D NumPy array representing the MRI image.
  • remove_empty_planes (bool, default: False): Whether to remove empty (non-brain) planes before processing.

compute_texture_values

Computes texture values for each 2D plane in the 3D image array.

Parameters

  • texture (str): The texture feature to compute (default is "contrast").

Returns

  • np.ndarray: An array of texture values for each 2D plane in the image.

extract_features

Extracts texture features from the MRI image by calculating statistical summaries for multiple texture metrics.

Parameters

  • textures (list[str], optional): A list of texture features to compute (e.g., 'contrast', 'energy').
    Defaults to ['contrast', 'dissimilarity', 'homogeneity', 'ASM', 'energy', 'correlation'].

Returns

A dictionary where keys represent texture feature names, and values represent the mean and standard deviation for each feature.