Radiomics

These features describe the distribution of individual voxel intensity values within a defined Region of Interest (ROI), independent of spatial relationships. They model the overall tissue density or signal intensity.

• Key Metrics: Mean, median, minimum, maximum, standard deviation, skewness, and kurtosis.
• Clinical Relevance: Skewness can indicate a shift in tissue homogeneity, while kurtosis reflects the peakedness of the histogram, often correlating with tumor grade.
• Example: A high standard deviation in a CT scan ROI suggests a highly heterogeneous tumor.

Shape features quantify the three-dimensional geometric properties of a segmented tumor volume. They are independent of the voxel intensity values and describe morphology.

• Key Metrics: Volume, surface area, sphericity, compactness, and maximum 3D diameter.
• Clinical Relevance: A high surface-to-volume ratio and low sphericity are often associated with more invasive, irregular tumor margins and poorer prognosis.
• Example: Sphericity is a dimensionless measure of how closely a tumor resembles a perfect sphere, with a value of 1.0 indicating a perfect sphere.

Texture features quantify the spatial inter-relationships between voxel intensities, capturing intratumoral heterogeneity patterns invisible to the human eye. They are derived from matrices that encode voxel pair arrangements.

• Gray-Level Co-occurrence Matrix (GLCM): Captures the frequency of adjacent voxel pairs with specific intensities, yielding metrics like contrast, correlation, energy, and homogeneity.
• Gray-Level Run-Length Matrix (GLRLM): Quantifies runs of consecutive voxels with the same intensity, capturing coarseness and directional patterns.
• Gray-Level Size Zone Matrix (GLSZM): Characterizes connected regions of identical voxel intensity, useful for identifying necrotic or cystic zones.

These features are extracted after applying mathematical filters to the original image, which amplifies specific frequency or structural patterns before quantification.

• Wavelet Decomposition: Applies high-pass and low-pass filters in 3D to decompose the image into different frequency sub-bands (e.g., LLL, HHH), revealing features at various scales.
• Laplacian of Gaussian (LoG): A band-pass filter that highlights edges and blobs of a specific size, enhancing features like spiculations.
• Clinical Relevance: Wavelet features often reveal subtle textural patterns that are strong independent predictors of distant metastasis and treatment response.

The computational process of converting medical images into a high-dimensional mineable feature space. This involves segmenting a Region of Interest (ROI) and calculating hundreds of quantitative descriptors.

• First-order statistics: Histogram-based metrics like mean, variance, skewness, and kurtosis of voxel intensities.
• Shape features: Geometric properties such as volume, surface area, sphericity, and compactness.
• Texture features: Gray-Level Co-occurrence Matrix (GLCM) and Gray-Level Run-Length Matrix (GLRLM) to quantify spatial heterogeneity.

A critical validation step to ensure radiomic features are capturing true biological signal rather than technical noise. Features are tested for stability against variations in acquisition and reconstruction.

• Test-retest analysis: Assessing feature consistency across repeated scans of the same patient.
• Inter-observer variability: Measuring the impact of different manual segmentations on feature values.
• Perturbation tests: Evaluating stability against image rotation, noise injection, and smoothing.

An advanced radiomics technique that partitions a tumor into distinct sub-regions, or habitats, based on clusters of similar voxel-level imaging characteristics. This reveals intratumoral heterogeneity invisible to the naked eye.

• Oxygenation mapping: Identifying hypoxic vs. well-perfused tumor niches.
• Cellular density: Distinguishing necrotic cores from highly proliferative rims.
• Treatment resistance: Pinpointing sub-volumes likely to survive targeted therapy.

The longitudinal analysis of radiomic feature changes over time, capturing the temporal dynamics of a disease or its response to treatment. It quantifies how texture and shape evolve between baseline and follow-up scans.

• Early response prediction: Detecting subtle textural changes before tumor shrinkage occurs.
• Pseudoprogression identification: Differentiating true progression from treatment-induced inflammation.
• Temporal delta features: Calculating the absolute and relative change in each feature between timepoints.

A statistical batch-effect correction method adapted from genomics to mitigate the center effect in multi-institutional radiomic studies. It harmonizes feature distributions across different scanner vendors and acquisition protocols.

• Location-scale adjustment: Modeling scanner-specific additive and multiplicative effects.
• Empirical Bayes estimation: Robustly estimating batch parameters, especially for small sample sizes.
• Preserved biological variance: Removing technical variability without destroying true pathological signal.

The integration of radiomic imaging phenotypes with genomic and transcriptomic data to establish non-invasive imaging surrogates for molecular characteristics.

• EGFR mutation prediction: Linking CT texture patterns to specific driver mutations in lung cancer.
• MGMT methylation status: Inferring promoter methylation from MRI texture in glioblastoma.
• Pathway enrichment: Correlating imaging features with activated biological signaling cascades.