- Atlas
- Medicine & Health
- Quantitative Imaging
Quantitative Imaging
Quantitative imaging is the branch of medical imaging that treats a picture as a measurement, not just a picture.
12 frameworks, from 1930 to today · Medical Imaging · Medicine & Health
- Densitometry 1930-1980
- Hounsfield scale 1972-Present
- Tracer kinetic modelling 1975-Present
- Diffusion-weighted magnetic resonance imaging 1985-Present
- Dual-energy X-ray absorptiometry 1985-Present
- Relaxometry 1985-Present
- Standardized uptake value 1985-Present
- Perfusion and diffusion quantification 1990-Present
- Perfusion scanning 1990-Present
- Quantitative imaging biomarker standardization (QIBA model) 2007-Present
- Radiomics 2012-Present
- Deep-learning reconstruction and analysis 2016-Present
Overview
Quantitative imaging is the branch of medical imaging that treats a picture as a measurement, not just a picture. A conventional radiologist reads a scan to say "this looks abnormal"; a quantitative imager asks how much of a physical or biological quantity is present, how reliably that number can be reproduced, and whether it means the same thing at a different hospital on a different scanner. The quantities range from the plainly physical—X-ray attenuation in Hounsfield units, bone mineral density in grams per square centimetre, tracer concentration in becquerels per millilitre—to the biological and statistical: blood flow through tissue, the apparent diffusion of water molecules, the fraction of a lesion's voxels that share a texture pattern123. What is at stake is comparability. A diagnosis can tolerate a somewhat fuzzy image; a measurement that will decide whether a tumour shrank, whether a drug worked, or whether a patient enters a trial cannot. Much of the history of this subfield is therefore the history of turning pictures into numbers that survive the journey between machines, sites, and readers.
From film density to calibrated attenuation
The earliest quantitative impulse in medical imaging came from radiography itself. Even on film, practitioners tried to relate the blackening of an image to the amount of X-ray absorbed, a practice known as densitometry4. Film density could be measured with a densitometer, and in the mid-twentieth century researchers used such measurements to estimate bone mineral content from plain radiographs—an early attempt to turn a shadow into a number. The approach was limited because film response, exposure, and scattered radiation all contaminated the reading, and because a radiograph collapses three dimensions onto one.
The decisive break came with computed tomography. In 1972 Godfrey Hounsfield and Allan Cormack's X-ray CT produced cross-sectional images whose pixel values were, by construction, calibrated to a physical property: the linear attenuation coefficient of tissue relative to water. Hounsfield defined a scale in which water is 0 and air is −1000, with dense bone running to roughly +1000 and beyond1. The Hounsfield unit (HU) was the first widely used quantitative imaging biomarker—a number with a defined physical meaning, comparable across scanners in principle. In practice early scanners drifted, and the field learned that even a calibrated unit needs quality control: phantom scans, calibration checks, and attention to beam hardening and reconstruction settings. The lesson that a physically defined unit is not automatically a stable measurement has recurred throughout the subfield's history.
Tracer kinetics and the standardized uptake value
Nuclear medicine approached quantification from a different direction5. Rather than measuring anatomy, it measured the distribution of radioactive tracers, and from the 1970s positron emission tomography (PET) made it possible to estimate tracer concentration in tissue over time. Dynamic PET studies with compartment models—mathematical descriptions of how a tracer moves between blood and tissue—could yield rates of blood flow, metabolism, and receptor binding. The best-known example is the measurement of glucose metabolism with fluorodeoxyglucose (FDG): kinetic modelling could estimate the metabolic rate of glucose in tissue, a genuinely quantitative physiological parameter.
Kinetic modelling, however, is demanding: it requires dynamic acquisition, arterial blood sampling, and complex analysis3. For routine clinical use a simpler index was needed. The standardized uptake value (SUV) emerged in the 1980s and 1990s as a practical surrogate—tissue activity concentration normalized by injected dose and body weight. The SUV is easy to compute from a static scan, but it is famously sensitive to body composition, plasma glucose, scan timing, reconstruction, and calibration. Its history is a case study in the tension between a convenient number and a trustworthy one: the SUV became ubiquitous in oncology despite well-documented variability, and much later work went into harmonizing it across scanners and sites.
Bone densitometry becomes a clinical measurement
Bone densitometry shows how a quantitative method can become a routine clinical test6. Dual-energy X-ray absorptiometry (DXA), developed in the 1980s, used two X-ray energies to separate bone from soft tissue and report areal bone mineral density in grams per square centimetre. DXA gave clinicians a number that predicted fracture risk and could be tracked over time, and it became the standard for diagnosing osteoporosis67. Its history also illustrates a recurring quantitative problem: areal density depends on bone size and on the software and reference population used, so results are not perfectly comparable between manufacturers. The field responded with standardization efforts and cross-calibration, themes that would grow into a movement.
Magnetic resonance: relaxometry, diffusion, and perfusion
MRI offered a far richer set of measurable tissue properties. Quantitative MRI relaxometry measures the relaxation times T1 and T2—the rates at which excited hydrogen nuclei return to equilibrium—which differ between tissues and can be mapped pixel by pixel89. Early quantitative MRI was slow and technically fragile, and much clinical MRI settled for qualitative contrast images. But relaxometry found durable niches, notably in myocardial tissue characterization, where T1 and T2 mapping help identify fibrosis and oedema, and in cartilage and brain research.
Two further quantitative techniques transformed clinical practice. Diffusion-weighted imaging, developed from the 1980s onward, measures how freely water molecules move; restricted diffusion in acute stroke or dense tumour tissue produces a measurable signal change, and the apparent diffusion coefficient (ADC) gives a number per voxel. Perfusion imaging measures blood flow through tissue, using tracer kinetics in CT and MRI or arterial spin labelling in MRI. Together, diffusion and perfusion quantification moved stroke and tumour assessment from "is there a lesion" toward "what kind of tissue is this, and how is it behaving"—numbers that guide treatment decisions.
The harmonization problem and the birth of a consortium
By the 2000s the subfield faced a structural problem. Each modality had produced its own quantitative indices, but the numbers were not portable. A CT attenuation value, an SUV, a DXA density, or an MRI relaxation time measured at one site could differ meaningfully from the same quantity measured elsewhere, because of scanner differences, acquisition protocols, reconstruction algorithms, and analysis software. Clinical trials and multicentre studies were undermined by this variability.
The response was organized standardization. The Quantitative Imaging Biomarkers Alliance (QIBA), convened under the Radiological Society of North America in the late 2000s, brought together clinicians, physicists, engineers, and industry to define profiles for specific biomarkers—what to measure, how to acquire and analyze it, and how to demonstrate that a measurement is repeatable and reproducible. QIBA's work established a vocabulary and a procedural template: a quantitative imaging biomarker is a measurement derived from an image that has a defined meaning and documented performance1011. The field also developed phantom-based validation, reference standards, and statistical frameworks for repeatability. This was not a single discovery but a shift in professional practice—from individual laboratories trusting their own numbers to a community agreeing on what a number means.
Radiomics and the statistical image
A parallel strand asked whether images contain quantitative information beyond what the eye extracts. Radiomics, which took shape around 2012, treats an image as a high-dimensional dataset: from each lesion, software computes hundreds of features describing shape, intensity distribution, and texture—statistical patterns of how voxel values vary in space. The hope was that these features, combined with machine learning, could predict diagnosis, prognosis, or treatment response12. Radiomics made explicit a problem that haunts all quantitative imaging: feature values depend heavily on acquisition and reconstruction parameters, segmentation choices, and software implementation. Multicentre reproducibility became the central methodological challenge, and the field responded with standardized feature definitions, open-source toolkits, and reporting guidelines. The related concept of quantitative imaging biomarkers in medical image analysis—using imaging measurements as endpoints in trials and clinical decision-making—grew alongside it1013.
Deep learning and learned reconstruction
The most recent turning point is the arrival of deep learning, which changed both what can be measured and how images are made14. Deep-learning reconstruction replaces or augments hand-designed reconstruction algorithms with neural networks trained on large datasets of measured and reference images. In CT, learned reconstruction can produce diagnostic-quality images from lower dose; in MRI, it accelerates acquisition; in PET, it improves signal-to-noise. The quantitative question is whether these learned images preserve the physical meaning of the underlying measurements—whether a Hounsfield unit or an SUV remains accurate after a network has "cleaned" the image. This is an active area of investigation, with phantom studies and task-based assessment used to check that learned reconstruction does not distort the quantities clinicians rely on.
Deep learning also extended radiomics into end-to-end learned features, and it enabled automated segmentation and measurement at scale. The subfield's present is thus a convergence: physically calibrated units from CT and nuclear medicine, physiological parameters from MRI and PET, statistical features from radiomics, and learned reconstruction and analysis—all facing the same demand for reproducibility and clinical validation.
Where the field stands
Today quantitative imaging is organized around a few durable commitments. Measurements are defined against physical or biological references where possible; performance is documented through repeatability and reproducibility studies; and standardization bodies produce profiles that specify acquisition, analysis, and validation. The main approaches in use descend directly from the history above: Hounsfield-unit CT and DXA for calibrated density, kinetic and SUV-based PET for tracer quantification, diffusion and perfusion MRI for tissue physiology, relaxometry for tissue characterization, radiomics for statistical image features, and deep-learning reconstruction for image formation. The open questions are the ones the history keeps raising: how to make a number mean the same thing across scanners and sites, how to validate a learned measurement against a physical ground truth, and how to demonstrate that a quantitative biomarker actually improves clinical decisions rather than merely being precise.
References
- Godfrey Hounsfield, Wikipedia
- Quantitative computed tomography, Wikipedia
- Standardized uptake value, Wikipedia
- Densitometry, Wikipedia
- Nuclear medicine, Wikipedia
- Dual-energy X-ray absorptiometry, Wikipedia
- Bone density, Wikipedia
- Relaxometry, Wikipedia
- Magnetic resonance imaging (History), Wikipedia
- Medical imaging (Use in pharmaceutical clinical trials), Wikipedia
- Biomarker (medicine), Wikipedia
- Radiomics, Wikipedia
- Imaging biomarker, Wikipedia
- Deep learning, Wikipedia
The frameworks of Quantitative Imaging
Each framework is a way of working the field has used: when it held, and the question it answers.
- 1930-1980
- 1972-Present
Hounsfield scale
What is the calibrated X-ray attenuation of this tissue?
- 1975-Present
Tracer kinetic modelling
How does a tracer move between blood and tissue over time?
- 1985-Present
Diffusion-weighted magnetic resonance imaging
How freely do water molecules diffuse in this tissue?
- 1985-Present
Dual-energy X-ray absorptiometry
What is the areal bone mineral density?
- 1985-Present
- 1985-Present
Standardized uptake value
How much tracer uptake is present relative to dose?
- 1990-Present
Perfusion and diffusion quantification
How much blood flows through and water diffuses in tissue?
- 1990-Present
- 2007-Present
Quantitative imaging biomarker standardization (QIBA model)
How can a quantitative biomarker be made reproducible across sites?
- 2012-Present
- 2016-Present
Deep-learning reconstruction and analysis
What can learned networks reconstruct and measure from images?
How the ideas connect
Each arrow runs from an earlier idea to the later one it shaped. The list below says what passed between them.
Hounsfield scale replaced Densitometry
Film densitometry tried to infer X-ray absorption from image blackening but was confounded by exposure and scatter. Hounsfield's CT replaced this with pixel values calibrated directly to linear attenuation coefficients, making attenuation a stable, physically defined number.
Carried by Godfrey Hounsfield; Allan Cormack.
Standardized uptake value grew out of Tracer kinetic modelling
Compartmental tracer kinetic modelling of dynamic PET yielded physiological rates but required arterial sampling and complex analysis. The standardized uptake value was devised as a simplified static-scan surrogate for tissue tracer uptake, trading kinetic rigour for clinical practicality.
Dual-energy X-ray absorptiometry grew out of Densitometry
Early radiographic densitometry estimated bone mineral content from film blackening. DXA refined this into a calibrated two-energy measurement of areal bone mineral density, inheriting the goal of turning a projection image into a bone quantity.
Quantitative imaging biomarker standardization (QIBA model) influenced by Hounsfield scale
The Hounsfield unit demonstrated that a physically calibrated image number still drifts across scanners and protocols, requiring phantom checks and quality control. QIBA generalized this lesson into formal profiles for establishing and documenting the reproducibility of quantitative imaging biomarkers.
Perfusion and diffusion quantification influenced by Tracer kinetic modelling
Tracer kinetic modelling supplied the compartmental mathematics for quantifying blood flow and tracer exchange. Perfusion and diffusion quantification in CT and MRI adapted these kinetic concepts to measure tissue perfusion and water mobility.
Perfusion scanning influenced by Tracer kinetic modelling
Perfusion scanning built on tracer kinetic principles, using contrast or labelled blood as a tracer and applying kinetic models to derive blood flow and volume. The modelling framework carried over from nuclear medicine into CT and MRI perfusion studies.
Quantitative imaging biomarker standardization (QIBA model) influenced by Standardized uptake value
The SUV's well-documented sensitivity to body composition, timing, reconstruction and calibration made it a paradigm case of an unstable quantitative index. QIBA's harmonization profiles addressed exactly this kind of cross-site variability for PET and other biomarkers.
Radiomics influenced by Quantitative imaging biomarker standardization (QIBA model)
QIBA established that a quantitative imaging biomarker needs defined meaning and documented repeatability, and promoted phantom validation and standardized reporting. Radiomics adopted this framework, making multicentre reproducibility and standardized feature definitions its central methodological concerns.
Deep-learning reconstruction and analysis influenced by Radiomics
Radiomics generated large sets of image-derived features and turned to machine learning for prediction, creating demand for automated feature extraction and analysis. Deep learning then supplied end-to-end learned representations and reconstruction, absorbing and extending the radiomic programme.
Quantitative imaging biomarker standardization (QIBA model) influenced by Dual-energy X-ray absorptiometry
DXA showed that areal bone density results differ between manufacturers and software versions, prompting cross-calibration and standardization efforts. This experience fed into the broader QIBA movement for reproducible quantitative biomarkers.
Quantitative imaging biomarker standardization (QIBA model) influenced by Perfusion and diffusion quantification
Diffusion and perfusion quantification produced numbers whose meaning depended on acquisition and analysis choices, underlining the need for agreed protocols. QIBA profiles later codified acquisition and analysis requirements for such biomarkers.
Beyond the field
Where Quantitative Imaging's ideas came from in other fields, and what they led to there: the strongest 12 of 38 links.
Tracer kinetic modelling influenced by Tracer Principle in Molecular Imaging
Tracer kinetic modelling applies the tracer principle, the use of a labelled substance to follow a physiological pathway, to derive rate constants from dynamic imaging data. The compartmental analysis of tracer behaviour descends directly from that principle.
Diffusion-weighted magnetic resonance imaging grew out of Magnetic Resonance Imaging in Medical Imaging
Diffusion-weighted MRI is a branch of MRI: it adds strong diffusion-sensitising gradients to a spin-echo sequence, building directly on the MRI hardware and pulse-sequence framework.
Carried by Denis Le Bihan.
Diffusion-weighted magnetic resonance imaging grew out of Physics of magnetic resonance imaging in Imaging Physics
Diffusion weighting follows from the physics of MRI: the Stejskal-Tanner relation between gradient pulses and signal attenuation is applied to the MR signal equation, so the technique is a direct extension of MRI physics.
Carried by Edward Stejskal; John Tanner.
Diffusion-weighted magnetic resonance imaging led to Diffusion MRI and Tractography in Neuroimaging
Diffusion MRI and tractography grew directly out of diffusion-weighted MRI: the pulsed-gradient measurement of water diffusion in tissue was extended to directional sampling and fibre tracking in the mid-1990s.
Carried by Peter Basser; Denis Le Bihan.
Relaxometry influenced by Nuclear magnetic resonance spectroscopy in Spectroscopy
MRI relaxometry descends from nuclear magnetic resonance spectroscopy: the measurement of spin-lattice and spin-spin relaxation times was developed in NMR and carried into imaging.
Carried by Felix Bloch; Edward Purcell.
Quantitative imaging biomarker standardization (QIBA model) grew out of Quantitative Molecular Imaging in Molecular Imaging
QIBA-style biomarker standardization grew directly out of quantitative molecular imaging's effort to make imaging measures reproducible across sites and scanners. It formalizes that field's quantitative aims into consensus profiles.
Deep-learning reconstruction and analysis influenced by Tomographic reconstruction in Imaging Physics
Deep-learning reconstruction builds on classical tomographic reconstruction, using its forward models and filtered-backprojection or iterative formulations as the structure networks learn to replace or accelerate. The mathematical problem it solves is the tomographic one.
Densitometry influenced by Radiography in Imaging Physics
Photographic film densitometry was developed to quantify the blackening of radiographs, so that exposure and film processing could be measured rather than judged by eye. The optical density concept and step-wedge calibration passed from radiography into densitometry.
Densitometry influenced by X-ray crystallography in Materials Science
X-ray crystallography's photographic films were quantified with microdensitometers, and the practice of measuring spot and line densities on diffraction films fed directly into densitometry as a technique.
Densitometry influenced by X-ray diffraction in Materials Characterization
Quantitative X-ray diffraction relied on measuring the blackening of photographic plates, and the microdensitometer developed for diffraction work became a standard densitometry instrument.
Densitometry influenced by Projectional radiography in Medical Imaging
Projectional radiography supplied the films whose optical density densitometry measured; sensitometry of radiographic film became a standard application of densitometry in radiology departments.
Tracer kinetic modelling influenced Receptor-Targeted Imaging in Molecular Imaging
Tracer kinetic modelling supplied the compartmental mathematics used to quantify receptor binding from dynamic imaging data. Receptor-targeted imaging adopted these kinetic models to estimate binding potential.

Quantitative Imaging is a planet in Nousfarer
Its city has a tower for each of these 12 frameworks. Land, walk in, and climb one concept at a time.
Land on Quantitative ImagingRelated fields
Fields its ideas are linked to
- Medical Imaging 11 links
- Molecular Imaging 7 links
- Imaging Physics 6 links
- Image Reconstruction 3 links
- Neuroimaging 3 links
- Medical Image Analysis 2 links
- Spectroscopy 1 link
- Materials Characterization 1 link
- Materials Science 1 link
- Imaging Informatics 1 link
- Stroke And Cerebrovascular Disease 1 link
- Nutritional Biochemistry 1 link
More in Medical Imaging
- Diagnostic Radiology
- Image Reconstruction
- Imaging Informatics
- Imaging Physics
- Interventional Radiology
- Medical Image Analysis
- Medical Imaging
- Molecular Imaging
- Nuclear Medicine
See where Quantitative Imaging sits on the Map of Human Knowledge