First-Order Statistics

The arithmetic average of all voxel intensity values within the volume of interest. Represents the central tendency of the tissue's signal.

• Clinical relevance: Distinguishes between hypo- and hyper-dense tissue regions
• Calculation: Sum of all voxel intensities divided by total voxel count
• Example: A mean Hounsfield Unit of 35 in a lung nodule suggests soft tissue density rather than calcification
• Limitation: Highly sensitive to outliers and does not capture heterogeneity

Variance measures the spread of intensity values around the mean, while standard deviation is its square root. These metrics quantify the dispersion of voxel intensities.

• Variance: Average of squared differences from the mean
• Standard deviation: Expressed in original intensity units for interpretability
• Clinical application: Higher standard deviation often correlates with tumor heterogeneity and potentially aggressive phenotypes
• Relationship: Together with mean, these form the basis for calculating higher-order moments like skewness and kurtosis

A measure of the asymmetry of the intensity histogram distribution around the mean value. Indicates whether the tail of the distribution extends toward higher or lower intensities.

• Positive skew: Tail extends toward higher intensities; mass of distribution concentrated on the left
• Negative skew: Tail extends toward lower intensities; mass concentrated on the right
• Zero skew: Symmetric distribution (e.g., normal distribution)
• Radiomic significance: Changes in skewness over time can indicate tissue transformation, such as necrosis development within tumors

A measure of the 'peakedness' or tailedness of the intensity histogram relative to a normal distribution. Quantifies the concentration of values around the mean versus the extremes.

• High kurtosis (leptokurtic): Sharp peak with heavy tails; indicates most voxels cluster tightly around the mean with some extreme outliers
• Low kurtosis (platykurtic): Flat peak with thin tails; indicates more uniform distribution of intensities
• Clinical example: High kurtosis in a glioblastoma region may indicate a homogeneous core with scattered necrotic or hemorrhagic foci
• IBSI compliance: Must be calculated as excess kurtosis (kurtosis - 3) per reference standards

A measure of the randomness or disorder in the distribution of voxel intensity values. Higher entropy indicates greater heterogeneity and unpredictability in the tissue signal.

• Calculation: Derived from the probability distribution of discretized intensity bins using Shannon's information theory formula
• Range: Minimum entropy occurs when all voxels share the same intensity; maximum occurs with uniform distribution across all bins
• Dependency: Highly sensitive to the number of intensity bins chosen during discretization
• Prognostic value: Elevated entropy in tumors has been associated with poorer treatment response and survival outcomes across multiple cancer types

A measure of the homogeneity of the intensity histogram. Quantifies how evenly voxel intensities are distributed across the defined bins.

• Maximum uniformity: All voxels fall within a single intensity bin (completely homogeneous)
• Minimum uniformity: Voxels are equally distributed across all bins
• Relationship to entropy: Uniformity is inversely related to entropy; high uniformity corresponds to low entropy
• Application: Used alongside entropy to characterize tissue organization; fibrotic tissue typically shows higher uniformity than necrotic or hemorrhagic regions

A first-order statistical measure of the randomness or disorder in the distribution of voxel intensity values within a region of interest. High entropy indicates a broad, flat histogram with many intensity values present in similar proportions, often associated with heterogeneous tissue. Low entropy suggests a narrow, peaked histogram where a few intensity values dominate, indicating homogeneous tissue. Mathematically, it quantifies the average information content required to encode the intensity distribution, making it a critical biomarker for assessing tumor cellular irregularity.

A first-order statistical measure of the asymmetry of the intensity histogram distribution around the mean value. A positive skew indicates the tail extends toward higher intensities (mass shifted to the left), while a negative skew indicates the tail extends toward lower intensities (mass shifted to the right). In oncological imaging, skewness captures subtle shifts in tissue density that may indicate necrotic core formation or edema, providing a quantitative descriptor of the overall shape of the voxel intensity distribution.

A first-order statistical measure of the peakedness or tailedness of the intensity histogram relative to a normal distribution. High kurtosis indicates a sharp central peak with heavy tails, suggesting most voxels cluster tightly around the mean with occasional extreme outliers. Low kurtosis indicates a flatter distribution. In radiomics, kurtosis helps differentiate cystic lesions (sharp peak at fluid density) from solid tumors with infiltrative margins, capturing the concentration of intensity values around the central tendency.

A second-order statistical method that quantifies texture by calculating how often pairs of pixels with specific values occur in a defined spatial relationship. Unlike first-order statistics, GLCM captures the spatial arrangement of intensities. Key derived features include:

• Contrast: Measures local intensity variation
• Homogeneity: Quantifies closeness to the diagonal, indicating uniformity
• Correlation: Assesses linear dependency between pixel pairs
• Energy: Measures textural uniformity (Angular Second Moment)

GLCM is foundational for characterizing tissue heterogeneity patterns invisible to histogram analysis alone.

The process of converting continuous image intensity values into a finite number of discrete bins, a critical pre-processing step for texture matrix calculation. Discretization directly impacts the size and statistical robustness of GLCM, GLRLM, and GLSZM matrices. Common strategies include:

• Fixed bin number: Divides the intensity range into a set number of bins (e.g., 64)
• Fixed bin width: Uses a constant intensity interval per bin

Inconsistent discretization across datasets is a primary source of non-reproducible radiomic features, making standardization essential per IBSI guidelines.

The computational process of removing unwanted technical variability from radiomic features caused by differences in scanner manufacturers, acquisition protocols, or reconstruction kernels. Without harmonization, batch effects can overwhelm biological signal. Key methods include:

• ComBat Harmonization: Adapts a Bayesian batch-effect correction framework from genomics to estimate and remove scanner-specific additive and multiplicative effects
• Normalization techniques: Z-score standardization or quantile normalization

Harmonization is essential for building robust radiomic signatures that generalize across multi-center clinical trials.