PyRadiomics

PyRadiomics extracts various radiomic features categorized into several classes. Most feature classes (except for Shape features) can be calculated on either the original image or a derived image obtained via filters.

Feature Classes:

• First Order Statistics (radiomics.firstorder): 19 features.
• Shape-based (3D) (radiomics.shape): 16 features. These are extracted from the label mask and are independent of gray value.
• Shape-based (2D) (radiomics.shape2D): 10 features.
• Gray Level Co-occurrence Matrix (GLCM) (radiomics.glcm): 24 features.
• Gray Level Run Length Matrix (GLRLM) (radiomics.glrlm): 16 features.
• Gray Level Size Zone Matrix (GLSZM) (radiomics.glszm): 16 features.
• Neighbouring Gray Tone Difference Matrix (NGTDM) (radiomics.ngtdm): 5 features.
• Gray Level Dependence Matrix (GLDM) (radiomics.gldm): 14 features.

Most features comply with the Imaging Biomarker Standardization Initiative (IBSI) definitions.

PyRadiomics generally adheres to IBSI (Image Biomarker Standardization Initiative) definitions, but there are notable differences in specific implementations that may cause variations in results:

• Binning (Fixed Bin Width): IBSI computes bin edges based on the resegmentation range or minimum intensities. PyRadiomics always uses bin edges equally spaced from 0, ensuring the lowest gray level is in the first bin.
• Resampling Alignment: IBSI aligns the resampling grid by the image center; PyRadiomics aligns it by the corner of the origin voxel. This can lead to differences in interpolated results and image/ROI sizes.
• Gray Value Rounding: PyRadiomics does not implement the IBSI practice of rounding resampled values to the nearest integer (to match original precision), as it considers the complexity-to-benefit ratio low.
• Mask Resampling: PyRadiomics forces mask resampling to use nearest neighbor interpolation to prevent incorrect re-assignment of label values.
• Kurtosis: IBSI calculates Excess Kurtosis (Kurtosis - 3). PyRadiomics calculates standard Kurtosis (which is +3 compared to IBSI).

In full-python mode, PyRadiomics allows reporting progress for GLCM and GLSZM matrix calculations. To enable this, set radiomics.progressReporter to a class object (not an instance) that satisfies these requirements:

1. Accepts an iterable and a keyword argument desc (string label).
2. Supports the context manager protocol (__enter__ and __exit__).
3. Is itself iterable (implements __iter__).

To intercept iteration for custom reporting, implement __iter__ to return self and define a __next__ method that calls the underlying iterable's __next__ method.

class MyProgressReporter(object):
    def __init__(self, iterable, desc=''):
        self.desc = desc
        self.iterable = iterable

    def __iter__(self):
        return self

    def __next__(self):
        nextElement = self.iterable.__next__()
        # Insert custom progress reporting code here
        return nextElement

    def __enter__(self):
        print (self.desc)
        return self

    def __exit__(self, exc_type, exc_value, tb):
        pass

To add a new feature class, create a separate module where the module name serves as the feature class name (e.g., tex.py becomes the tex feature class). The class must inherit from radiomics.base.RadiomicsFeaturesBase.

Key requirements:

• Initialization: Call super().__init__(inputImage, inputMask, **kwargs) in your __init__ method.
• Feature Methods: Define individual features using the signature get[Name]FeatureValue(self), which must return a scalar value.
• C Extensions: If using C extensions for matrix calculations, implement a _calculateMatrix function that returns a numpy array and assign it to a class variable named P_[Name] (where [Name] is the module name).
• Logging: Use self.logger (a child logger of the radiomics namespace) for all log messages.
• Documentation: Docstrings are required at both the class and individual feature levels.

from radiomics import base

class Radiomics[Name](base.RadiomicsFeaturesBase):
    """
    Feature class docstring
    """

    def __init__(self, inputImage, inputMask, **kwargs):
        super(Radiomics[Name], self).__init__(inputImage, inputMask, **kwargs)
        # Feature class specific init

    def get[Feature]FeatureValue(self):
        """
        Feature docstring
        """
        # value = feature calculation using member variables of RadiomicsFeatureBase and this class.
        return [value]