Gray-Level Size Zone Matrix (GLSZM)

Unlike the Gray-Level Run Length Matrix (GLRLM) which requires computation across multiple directional angles, the GLSZM is inherently rotationally invariant. It counts the number of connected voxels sharing the same gray level in a 3D neighborhood regardless of their spatial orientation. This property makes it uniquely suited for analyzing amorphous tumor structures where a dominant growth direction does not exist, ensuring consistent feature values irrespective of patient positioning or scan alignment.

A zone is defined as a connected group of voxels that all share the exact same discretized gray-level intensity. Connectivity is typically defined using a 26-connected neighborhood in 3D space. The resulting matrix P(i,j) stores the frequency of a zone with gray level i and size j. Key structural implications:

• Diagonal entries represent small, isolated homogeneous pockets
• Off-diagonal entries capture large, contiguous regions of uniform intensity
• The matrix inherently captures both micro-heterogeneity (many small zones) and macro-heterogeneity (few large zones)

Zone Percentage measures the homogeneity of zone sizes across the entire region of interest. It is calculated as the fraction of the total number of zones that correspond to the most prevalent zone size. A high ZP value indicates a coarse, uniform texture where one zone size dominates the matrix. A low ZP suggests a heterogeneous texture with a wide distribution of zone sizes. This feature is particularly sensitive to necrotic cores in tumors, where large homogeneous regions of low intensity form.

Large Zone Emphasis quantifies the distribution of large homogeneous areas within the image. It is calculated by weighting zone counts by the square of their size, making it highly sensitive to the presence of expansive, uniform regions. Clinical relevance includes:

• High LZE: Indicates broad areas of uniform cellularity, often seen in well-differentiated tumors or cystic components
• Low LZE: Suggests a finely textured, highly heterogeneous tissue architecture
• LZE is a critical differentiator in distinguishing invasive ductal carcinoma from benign fibroadenomas in breast MRI

Gray-Level Non-Uniformity measures the variability of intensity values across the image by assessing the distribution of zones among different gray levels. A low GLN value signifies that zones are evenly distributed across all intensity bins, indicating isotropic intensity distribution. A high GLN value means certain gray levels dominate the zone count, suggesting intensity heterogeneity. This feature is instrumental in quantifying metabolic activity in PET/CT fusion imaging, where standardized uptake value (SUV) discretization reveals functional tumor sub-volumes.

Small Zone Emphasis is the inverse counterpart to LZE, designed to highlight fine, granular textures. It weights smaller zones more heavily in its calculation, making it a sensitive detector of micro-calcifications and early fibrotic changes. Key characteristics:

• Elevated SZE correlates with trabecular bone deterioration in osteoporosis screening
• In lung CT, high SZE values are associated with ground-glass opacity patterns indicative of early-stage adenocarcinoma
• SZE is mathematically normalized to ensure comparability across different ROI volumes