Glossary

Radiomics

The high-throughput extraction and analysis of quantitative features from medical images to create mineable data for decision support.

Get in touch Learn more

Data engineer managing feature store on laptop, feature definitions visible, casual data engineering session.

HIGH-THROUGHPUT IMAGING BIOMARKER DISCOVERY

What is Radiomics?

Radiomics is the high-throughput extraction and analysis of quantitative features from medical images to create mineable data for decision support.

Radiomics is the automated, high-throughput conversion of standard-of-care medical images (CT, MRI, PET) into high-dimensional, mineable data. It operates on the hypothesis that quantitative image features—capturing tumor intensity, shape, and texture—reflect underlying pathophysiology invisible to the naked eye. This process involves region of interest (ROI) delineation, feature extraction using mathematical algorithms like Gray-Level Co-occurrence Matrix (GLCM) and wavelet transforms, and subsequent statistical modeling to build predictive radiomic signatures.

The core engineering challenge lies in feature harmonization to remove scanner-induced variability using methods like ComBat harmonization, ensuring cross-institutional reproducibility. By linking these extracted features to genomic profiles or clinical outcomes, radiomics serves as a non-invasive, repeatable virtual biopsy. The field is governed by standards from the Image Biomarker Standardisation Initiative (IBSI) to ensure that robust feature selection and dimensionality reduction techniques produce clinically actionable, validated biomarkers.

QUANTITATIVE IMAGING PHENOTYPES

Core Feature Classes in Radiomics

Radiomic features are broadly categorized into distinct mathematical families, each capturing a different aspect of the underlying tissue phenotype. These classes form the vocabulary for translating medical images into high-dimensional mineable data.

First-Order Statistics (Histogram)

These features describe the distribution of individual voxel intensities within the Region of Interest (ROI) without any consideration of spatial relationships. They are derived from the image histogram and quantify properties like the central tendency, dispersion, and shape of the intensity distribution.

Key Metrics: Mean, Median, Minimum, Maximum, Standard Deviation, Skewness (asymmetry), Kurtosis (peakedness), and Entropy (randomness).
Clinical Relevance: High entropy often correlates with heterogeneous tumors, while skewness can indicate the presence of necrotic cores or enhancing rims.
Vulnerability: Highly sensitive to intensity discretization and normalization parameters.

Common IBSI Features

Shape and Morphology Features

These descriptors quantify the three-dimensional geometric properties of the segmented Volume of Interest (VOI). They are independent of the intensity values and rely solely on the binary mask defining the structure's boundary.

2D vs 3D: Includes maximum 2D diameter for a single slice, but the most robust features are volumetric (3D).
Key Metrics: Sphericity (how closely the shape resembles a sphere), Surface-to-Volume Ratio, Compactness, and Maximum 3D Diameter.
Application: A low sphericity value often indicates an infiltrative, irregular tumor margin, which is a hallmark of malignancy in many cancer types.

Required Geometry

Texture Matrices (Second & Higher Order)

Texture analysis quantifies the spatial arrangement of voxel intensities, capturing inter-pixel relationships that define concepts like coarseness, contrast, and directionality. These matrices form the core of handcrafted radiomics.

Gray-Level Co-occurrence Matrix (GLCM): Calculates the probability of pairs of pixels with specific values occurring at a defined distance and angle. Produces Contrast, Homogeneity, and Correlation.
Gray-Level Run Length Matrix (GLRLM): Counts consecutive pixels with the same gray level in a specific direction. Produces Run Emphasis and Run Percentage.
Gray-Level Size Zone Matrix (GLSZM): Quantifies connected regions of identical gray levels, independent of direction. Excellent for capturing intratumoral heterogeneity.

GLCM, GLRLM, GLSZM, NGTDM

Core Matrix Families

Filter-Based and Transform Features

These features are extracted after applying specific mathematical filters to the original image, allowing the quantification of patterns at different spatial scales or frequency domains that are invisible to the naked eye.

Wavelet Transform: Decomposes the image into high- and low-frequency sub-bands (e.g., LHL, HHH) in the x, y, and z dimensions. Texture features are then recalculated on these decomposed images.
Laplacian of Gaussian (LoG): An edge-detection filter that applies Gaussian smoothing before computing the Laplacian. Varying the sigma value highlights structures of different sizes (blobs vs. fine edges).
Impact: Dramatically expands the feature space, often generating thousands of derived features from a single VOI.

Wavelet Decompositions

Deep Radiomic Features

Unlike handcrafted features defined by explicit mathematical formulas, Deep Radiomics utilizes Convolutional Neural Networks (CNNs) to automatically learn hierarchical feature representations directly from the image data.

Mechanism: Features are extracted from the bottleneck or final convolutional layers of a pre-trained network (e.g., ResNet, VGG) applied to 2D slices or 3D patches.
Advantage: Can capture abstract, high-level semantic patterns that are difficult to engineer manually.
Challenge: These features lack the explicit mathematical interpretability of GLCM or shape features, often functioning as a 'black box' and requiring robust model explainability techniques for clinical validation.

Transfer Learning

Common Strategy

Delta-Radiomic Features

This class captures the temporal dynamics of disease by quantifying the net change in radiomic features between two time points, typically pre- and post-treatment.

Calculation: Defined as (Feature_Post - Feature_Pre) / Feature_Pre. It measures the relative variation in texture or morphology over time.
Clinical Utility: A strong predictor of treatment response and survival in oncology. For example, a significant decrease in tumor Entropy post-chemotherapy often indicates a favorable pathological response.
Requirement: Demands strict test-retest reproducibility and robust feature harmonization to ensure that measured changes are biological rather than technical noise.

Temporal

Feature Domain

RADIOMICS CLARIFIED

Frequently Asked Questions

Concise answers to the most common technical questions about high-throughput medical image mining and quantitative feature extraction.

Radiomics is the high-throughput extraction and analysis of quantitative features from medical images to create mineable data for decision support. The process works by converting standard-of-care medical images (CT, MRI, PET) into high-dimensional data through a multi-step pipeline: image acquisition, region of interest (ROI) delineation, preprocessing (voxel resampling and intensity discretization), feature extraction (first-order statistics, texture matrices like GLCM and GLRLM, and shape descriptors), and predictive modeling. The core hypothesis is that medical images contain latent information about tissue pathophysiology—such as tumor heterogeneity and genetic expression—that is not visible to the naked eye but can be quantified algorithmically and correlated with clinical endpoints like survival or treatment response.

Enabling Efficiency, Speed & Accuracy

Intelligent Analysis, Decision & Execution

We build AI systems for teams that need search across company data, workflow automation across tools, or AI features inside products and internal software.

Talk to Us

Search across company data

Give teams answers from docs, tickets, runbooks, and product data with sources and permissions.

Useful when people spend too long searching or get different answers from different systems.

Enterprise searchRAGPermissions

Automate internal workflows

Use AI to route work, draft outputs, trigger actions, and keep approvals and logs in place.

Useful when repetitive work moves across multiple tools and teams.

AI agentsWorkflow automationGovernance

Add AI to products and internal tools

Build assistants, guided actions, or decision support into the software your team or customers already use.

Useful when AI needs to be part of the product, not a separate tool.

AI integrationDecision supportModel routing

About the author

Prasad Kumkar

CEO & MD, Inference Systems

Prasad Kumkar is the CEO & MD of Inference Systems and writes about AI systems architecture, LLM infrastructure, model serving, evaluation, and production deployment. Over 5+ years, he has worked across computer vision models, L5 autonomous vehicle systems, and LLM research, with a focus on taking complex AI ideas into real-world engineering systems.

His work and writing cover AI systems, large language models, AI agents, multimodal systems, autonomous systems, inference optimization, RAG, evaluation, and production AI engineering.

Limited slotsGet a Free AI Consultation

How We Work

Custom AI workflows for your Business

One-fit-all AI don't work for modern businesses. At Inferensys, we aim to understand your business & custom requirements; which we use to define most efficient agentic workflows, the data, and the tools for your business.

Talk to Us

Radiomics

What is Radiomics?