Table 1. Description of radiographic features and filters.
Individual descriptions are given for each group and parameter or feature.
| Type | Group | Feature / Parameter | Description |
|---|---|---|---|
| Radiographic features | Semantic | Intratumoral heterogeneity | Heterogeneity in hyperintensity of MRI signal throughout tumor |
| Multifocality | Non-contiguous growth of tumor | ||
| Midline shift | Shift of the brain past midline | ||
| Sinus invasion | Presence of venous sinus invasion | ||
| Necrosis / Hemorrhage | Presence of necrosis or hemorrhage | ||
| Mass effect | Shift in normal brain parenchyma due to tumor | ||
| Cystic component | Fluid filled cysts within the tumor | ||
| Bone invasion | Appearance of tumor invading the skull | ||
| Hyperostosis | Bony overgrowth adjacent to tumor | ||
| Spiculation | Irregularities in tumor shape and border | ||
| Radiomic | Median | Median voxel intensity value | |
| Mean | Mean voxel intensity value | ||
| Minimum | Minimal voxel intensity value | ||
| Skewness | Describes the shape of a probability distribution of the voxel intensity histogram | ||
| Spherical Disproportion (SD) | How different is the tumor is to a sphere with a similar volume | ||
| Cluster Prominence (CP) | Sensitive to flat zones (area of similar intensity) | ||
| Difference Entropy (DE) | Complexity of the pattern (high entropy for high number of unique patterns) | ||
| Inverse Difference Normalized (IDN) | Sensitive to homogeneity in the tumor | ||
| Run Length Non-uniformity (RLN) | Measure of heterogeneity | ||
| Short Run Low Gray-Level Emphasis (SRLGLE) | Measure of heterogeneity sensitive to low intensity pattern | ||
| High Intensity Large Area Emphasis (HILAE) | Sensitive to flat zones with high intensity voxels (e.g. areas of hemorrhage) | ||
| Low Intensity Large Area Emphasis (LILAE) | Sensitive to flat zones with low intensity voxels (e.g. areas of necrosis) | ||
| Low Intensity Small Area Emphasis (LISAE) | Sensitive to small flat zones with low intensity voxels | ||
| Filters | Wavelet | High (L), Low (L) | Wavelet filters decompose images by high (increase details) and low (smooth image, leaving general shape) for every spatial component (x,y,z) |
| LoG | Sigma (σ) | Laplacian of Gaussian is a filter that highlights textures using a variable size radius (σ). Depending on the radius (from 0.5mm to 5mm with 0.5 increment), it emphasizes image textures from fine to coarse. |