PyRadiomics

Implements a complete panel of standardized radiomic features as defined by the Image Biomarker Standardisation Initiative (IBSI). Extracts First-Order Statistics, Shape Features (3D and 2D), and multiple texture matrices including GLCM, GLRLM, GLSZM, GLDM, and NGTDM.

• Extracts over 100 distinct feature classes from a single Region of Interest
• Supports both 2D slice-based and full 3D volumetric analysis
• Provides wavelet and LoG filtered image decompositions for multi-scale texture analysis

Natively ingests DICOM and NIfTI formats, the two primary standards in medical imaging. Handles complex DICOM tag parsing to automatically extract voxel spacing, slice thickness, and patient orientation.

• Reads DICOM series directories and single NIfTI volumes
• Automatically applies spatial resampling to ensure isotropic voxel dimensions
• Preserves metadata provenance throughout the extraction pipeline

Enforces a deterministic preprocessing workflow to ensure feature reproducibility across institutions. Applies intensity discretization using fixed bin width or fixed bin count strategies before texture matrix calculation.

• Configurable intensity normalization with scale and outlier filtering
• Built-in resampling to user-defined isotropic voxel spacing
• Optional image normalization to zero mean and unit variance prior to wavelet decomposition

Engineered for high-throughput research workflows processing large patient cohorts. Supports multi-threaded extraction to parallelize feature computation across multiple CPU cores.

• Processes thousands of imaging studies in a single execution
• Generates structured CSV output for direct ingestion into statistical modeling pipelines
• Integrates seamlessly with SimpleITK for accelerated image filtering operations

Feature definitions and computation algorithms are validated against the Image Biomarker Standardisation Initiative consensus reference values. This ensures that extracted features are mathematically equivalent to those produced by other IBSI-compliant platforms.

• Enables multi-center study harmonization
• Provides transparent documentation of all feature formulas and computational steps
• Supports ComBat harmonization workflows to correct for scanner-induced batch effects

Applies a bank of wavelet and Laplacian of Gaussian (LoG) filters to the original image before feature extraction. This decomposes the image into multiple frequency sub-bands, capturing textural patterns at different spatial scales.

• Wavelet decompositions across all combinations of high-pass and low-pass filtering in 3D
• Configurable LoG sigma values for multi-scale edge enhancement
• Features extracted from each filtered image are labeled with filter parameters for traceability

A critical preprocessing step that bins continuous voxel intensity values into a finite number of discrete gray levels. PyRadiomics implements fixed bin number and fixed bin width strategies. This quantization directly impacts the reproducibility of all subsequent texture matrices.

• Fixed Bin Width: Bins are defined by a constant Hounsfield Unit or intensity interval.
• Fixed Bin Number: The intensity range is divided into a specified number of bins.
• Directly controls the granularity of GLCM and GLRLM calculations.

An independent international collaboration providing consensus-based reference values and standardized nomenclature for radiomic feature computation. PyRadiomics benchmarks its feature calculations against the IBSI digital phantom to ensure mathematical compliance and reproducibility across institutions.

• Validates First-Order, Shape, and Texture feature outputs.
• Provides a common language to eliminate ambiguity in feature naming.
• Essential for multi-center clinical trial harmonization.

PyRadiomics applies Wavelet Transform and Laplacian of Gaussian (LoG) filters to the original image before feature extraction. This isolates textural information at different spatial scales and frequency bands.

• Wavelet: Decomposes the image into high-pass (H) and low-pass (L) sub-bands (e.g., LHL, HHH) in 3D.
• LoG: Highlights regions of rapid intensity change; the sigma value defines the scale of the detected edges.
• Enables extraction of localized textural features invisible to standard histogram analysis.

Morphological descriptors quantifying the three-dimensional geometric properties of a segmented Region of Interest (ROI). These features are independent of intensity values and rely solely on the mask.

• Sphericity: Measures how closely the tumor approximates a perfect sphere.
• Maximum 3D Diameter: The largest pairwise Euclidean distance between surface mesh vertices.
• Surface Area to Volume Ratio: Indicates irregularity and invasiveness of the lesion margin.

A second-order statistical method that quantifies texture by calculating the frequency of specific pairs of pixel intensities occurring at a defined spatial offset. PyRadiomics computes the GLCM in 3D by averaging across 13 directions.

• Contrast: Measures local intensity variation; high contrast implies a heterogeneous texture.
• Correlation: Assesses the linear dependency of gray levels on neighboring voxels.
• Energy (Angular Second Moment): Measures textural uniformity and orderliness.

A statistical batch-effect correction method adapted from genomics to remove non-biological technical variance in radiomic features. PyRadiomics features can be harmonized post-extraction to mitigate the scanner effect.

• Adjusts for systematic differences between CT scanner manufacturers or acquisition protocols.
• Preserves biological variability while removing technical noise.
• Critical for building robust radiomic signatures in multi-center retrospective studies.