Image Biomarker Standardisation Initiative (IBSI)

IBSI distributes a standardised digital phantom—a synthetic 3D image dataset with known geometric and textural properties. Research groups use this phantom to:

• Benchmark their extraction software against reference values
• Detect implementation errors in feature calculation code
• Quantify numerical precision across different computing environments

The phantom contains distinct zones designed to exercise specific feature families, including homogeneous regions, edge transitions, and repeating patterns.

IBSI defines a step-by-step processing pipeline that must be applied before feature extraction to ensure cross-study comparability:

• Intensity discretisation: Fixed bin number (FBN) and fixed bin size (FBS) methods with recommended parameters
• Interpolation: Preferred algorithms for resampling to isotropic voxel spacing
• Intensity rescaling: Mapping CT values to standardised Hounsfield Unit ranges
• Re-segmentation: Absolute and relative thresholding strategies for excluding non-tissue voxels

Each step includes explicit parameter recommendations to reduce investigator degrees of freedom.

IBSI establishes minimum reporting standards for radiomic studies to enable independent replication:

• Feature family and name must be reported using IBSI nomenclature
• Discretisation method and parameters must be explicitly stated
• Software version and implementation must be cited
• Deviations from IBSI standards must be documented and justified

These guidelines align with the broader Radiomics Quality Score (RQS) framework and support regulatory submissions.

IBSI organises features into a hierarchical taxonomy of distinct families:

• Morphological: Volume, surface area, sphericity, compactness
• First-order statistics: Mean, variance, skewness, kurtosis, entropy
• Texture matrices: GLCM, GLRLM, GLSZM, NGTDM, GLDM
• Filter-based: Wavelet decompositions, Laplacian of Gaussian band-pass filtering

Each category has defined mathematical properties and recommended use cases for characterising different aspects of tissue heterogeneity.

The computational process of removing unwanted technical variability from radiomic features caused by differences in scanner models, acquisition protocols, or reconstruction kernels. IBSI provides the standardized feature definitions, but harmonization techniques like ComBat are required to align feature values across cohorts. Without harmonization, batch effects can dominate biological signal, rendering multi-center radiomic signatures non-reproducible.

The process of converting continuous image intensity values into a finite number of discrete bins, a critical pre-processing step for texture matrix calculation. IBSI mandates fixed bin number (FBN) and fixed bin size (FBS) as the two standardized discretization approaches. The choice of bin width directly impacts texture feature values, making IBSI's strict discretization guidelines essential for cross-study comparability.

A dimensionality reduction strategy that identifies and retains only radiomic features demonstrating high stability against test-retest and inter-observer variability. IBSI's reference standards enable researchers to first verify which features are computationally reproducible before assessing biological relevance. Features with intraclass correlation coefficient (ICC) below 0.75 are typically discarded to ensure generalizable radiomic signatures.

The use of deep convolutional neural networks to automatically learn hierarchical feature representations directly from medical images, bypassing handcrafted feature engineering. While IBSI focuses on engineered features with explicit mathematical definitions, deep radiomics raises new standardization challenges around architecture reproducibility and weight initialization. Emerging efforts seek to extend IBSI-like benchmarking to learned feature extractors.

A composite biomarker consisting of a specific set of weighted radiomic features combined via a mathematical model to predict a clinical endpoint such as overall survival or treatment response. IBSI compliance ensures that the individual features composing a signature are calculated identically across institutions. Signature validation requires demonstrating both IBSI-compliant feature extraction and robust performance on external cohorts.