- Atlas
- Medicine & Health
- Medical Imaging
Medical Imaging
Medical imaging is the practice of making the inside of a living body visible without cutting it open.
13 frameworks, from 1895 to today · Medical Imaging · Medicine & Health
- Projectional radiography 1895-Present
- Fluoroscopy 1896-Present
- Medical ultrasound 1950-Present
- Nuclear medicine 1950-Present
- Image Reconstruction and Computational Imaging 1970-Present
- CT scan 1972-Present
- Magnetic Resonance Imaging 1973-Present
- Digital Radiography and Imaging Informatics 1980-Present
- Computer-aided diagnosis 1985-Present
- Molecular Imaging 1995-Present
- Hybrid and Multimodality Imaging 2000-Present
- Radiomics 2010-Present
- Deep Learning in Medical Imaging 2012-Present
Overview
Medical imaging is the practice of making the inside of a living body visible without cutting it open1. Its practitioners—radiologists, radiographers, nuclear physicians, sonographers, medical physicists, and engineers—build and operate machines that map some physical property of tissue onto a picture a clinician can read. What is at stake is diagnosis and treatment: whether a tumour is present, where a fracture runs, whether blood still reaches the heart muscle, whether a treatment is working. The history of the field is a history of finding new physical signals to exploit and new mathematics to turn those signals into images.
Shadows on Film: The Discovery of X-rays and Clinical Radiography
In November 1895 Wilhelm Röntgen, working with a cathode-ray tube in Würzburg, noticed that a screen coated with barium platinocyanide glowed even when shielded from the tube. He called the unknown radiation X-rays and within weeks produced a radiograph of his wife's hand showing the bones and her ring. The physical basis was differential attenuation: dense tissue such as bone absorbs more of the rays than soft tissue or air, so a beam passing through the body casts a shadow whose contrast encodes internal structure2.
Clinical radiography spread with extraordinary speed. By 1896 physicians in Europe and North America were imaging fractures and foreign bodies, and within a few years the fluoroscope—an X-ray tube behind a fluorescent screen—let a clinician watch motion in real time, useful for swallowing studies and later for guiding procedures. The early decades were dominated by the practical problems of the technique: protecting patients and operators from radiation, standardising exposure, and improving contrast with contrast agents. Barium sulphate suspensions, introduced in the early twentieth century, made the gastrointestinal tract visible; iodine-based compounds later did the same for the urinary tract and blood vessels. Projection radiography—the flat shadow image—remained the workhorse of medical imaging for most of the century.
Seeing in Slices: Computed Tomography
Projection radiography collapses a three-dimensional body onto a two-dimensional shadow, so structures overlap. The mathematical answer had been known since 1917, when Johann Radon showed that a function could be reconstructed from its line integrals—its projections at many angles34. The obstacle was computational: reconstructing a slice by hand was impractical.
Godfrey Hounsfield, an engineer at EMI, built the first practical X-ray computed tomography (CT) scanner and published results in 1972; Allan Cormack had independently derived the reconstruction mathematics in the 1960s. The scanner rotated an X-ray source and detector around the patient and measured attenuation along hundreds of paths, then a computer reconstructed a cross-sectional image. The first clinical scans were of the head, where CT could distinguish grey and white matter, blood, and tumour—something no radiograph could do. Hounsfield and Cormack shared the 1979 Nobel Prize in Physiology or Medicine13.
CT transformed diagnosis and drove a generation of reconstruction research. The filtered back-projection algorithm, which back-projects measured values along their paths and applies a filter to sharpen the result, became standard5. Later, spiral (helical) scanning, in which the patient moves continuously through a rotating gantry, made fast whole-body studies routine, and multi-detector rows pushed scan times down far enough to image a beating heart.
From Film to Numbers: Digital Radiography and the Informatics Turn
Radiography was photographic for most of its history: a film emulsion darkened where X-rays struck it. Digital radiography replaced film with electronic detectors—photostimulable phosphor plates in computed radiography from the 1980s, and flat-panel detectors that converted X-rays directly into electrical signals. The gain was not only convenience: digital images could be windowed, transmitted, stored, and processed. Picture archiving and communication systems (PACS) let hospitals replace film libraries with networks, and the radiology report began to be treated as structured data rather than prose. This informatics layer is easy to overlook but it is what made later computational methods possible, because images became arrays of numbers available to software.
Watching Function: Nuclear Medicine and Molecular Imaging
Radiography and CT show anatomy—where things are and how dense they are. Nuclear medicine shows physiology. In 1927 Hermann Blumgart and Otto Yens used a radioactive tracer to measure blood circulation time in humans, and by the 1950s the gamma camera, developed by Hal Anger, could form an image of where a radionuclide had accumulated. The principle is emission: a radiopharmaceutical is administered, it concentrates in particular tissue according to its biochemistry, and the emitted gamma rays are detected. Bone scans, thyroid imaging, and myocardial perfusion studies followed.
Positron emission tomography (PET), developed from the 1950s onward and refined through the 1970s, detects pairs of gamma rays produced when a positron annihilates with an electron, giving far greater sensitivity and allowing tracers such as fluorine-18 labelled glucose (FDG) to map metabolic activity—cancer cells and active brain regions take up glucose avidly. Molecular imaging generalises the idea: image the molecules and processes of disease, not just the shape of organs. The field's ambition is to see a receptor, an enzyme, or a gene's activity in a living patient.
Sound and Magnetism: Ultrasound and MRI
Two imaging modalities arrived that used neither ionising radiation nor injected radioactivity. Diagnostic ultrasound, developed clinically from the 1950s, sends high-frequency sound into the body and builds an image from echoes; because it is real-time and portable, it became central to obstetrics, cardiology, and emergency medicine. Magnetic resonance imaging (MRI) came from nuclear magnetic resonance, the physics of how atomic nuclei in a magnetic field absorb and re-emit radio waves6. Paul Lauterbur showed in 1973 that magnetic field gradients could encode spatial position into the NMR signal, and Peter Mansfield developed the fast imaging methods that made MRI practical; they shared the 2003 Nobel Prize in Physiology or Medicine67. MRI's great advantage is soft-tissue contrast without ionising radiation, and it can be tuned to many tissue properties—water content, fat, blood flow, diffusion of molecules—by changing the pulse sequence1.
Reconstructing the Invisible: Computational Imaging
Every slice-based modality depends on reconstruction: turning indirect measurements into a picture. The field matured from analytic methods such as filtered back-projection to iterative reconstruction, which models the physics of the measurement and the statistics of noise and solves for the image most consistent with the data. Iterative methods allow lower dose and fewer measurements, at the cost of computation. Compressed sensing, applied to MRI in the mid-2000s, showed that images could be reconstructed from far fewer samples than classical theory demanded by exploiting the fact that medical images are sparse in some transform. This line of work—image reconstruction and computational imaging—is where physics, mathematics, and engineering meet, and it is the reason modern scanners can image faster and with less radiation than their predecessors.
Combining Modalities: Hybrid Imaging
Each modality sees something the others miss, so practitioners learned to fuse them. PET and CT were combined in a single scanner in the early 2000s, so that the functional signal of PET could be overlaid on the anatomical map of CT acquired in the same session, with the patient in the same position8. PET/CT became standard in oncology. PET/MRI followed, offering functional imaging with MRI's soft-tissue contrast and lower radiation. Hybrid and multimodality imaging is now a design principle: register images from different devices, or build devices that acquire several signals at once, and present them as one aligned dataset.
The Machine Learns: Computer-Aided Detection and Deep Learning
From the 1980s, researchers tried to have computers find lesions—microcalcifications in mammograms, lung nodules in chest radiographs—under the name computer-aided detection (CAD). Early systems used handcrafted features: engineers specified what a nodule looked like (roundness, edge gradient, density) and a classifier searched for those patterns. Results were mixed, and CAD's clinical reputation suffered from high false-positive rates.
The turning point was deep learning. Convolutional neural networks, trained on large labelled image sets, learn their own features rather than being told them9. A landmark demonstration came in 2012 when a deep network won the ImageNet recognition challenge by a wide margin; medical imaging followed within a few years, with networks matching or exceeding radiologists on narrow tasks such as detecting diabetic retinopathy or classifying skin lesions in research settings. The field now has a large literature on deep learning for detection, segmentation, and diagnosis, and the practical questions have shifted to validation, generalisation across hospitals, and regulation.
A related development is radiomics: extracting large numbers of quantitative features from images—texture, shape, intensity statistics—and correlating them with outcomes such as tumour genotype or treatment response10. Radiomics and computer-aided detection share the goal of mining images for information the eye cannot reliably extract, and both now lean heavily on machine learning.
Where the Field Stands
Today the main modalities coexist and are chosen by the clinical question: projection and digital radiography for quick skeletal and chest work, CT for fast cross-sectional anatomy and trauma, MRI for soft tissue and function, ultrasound for real-time and bedside imaging, nuclear medicine and PET for metabolism and molecular targets, and hybrid scanners for combined anatomical and functional maps1611. Reconstruction has become iterative and model-based, and deep learning is entering the reconstruction pipeline itself as well as the reading room. The open questions are the ones the history has been circling: how to see smaller and earlier, how to lower dose and cost, how to make computational tools that generalise beyond the data they were trained on, and how to prove that a new image or a new algorithm actually improves what happens to patients.
References
- Medical imaging (introduction, History, Magnetic resonance imaging, Radiography, Ultrasound, Tomography), Wikipedia
- Radiography, Wikipedia
- CT scan (History, introduction), Wikipedia
- Tomographic reconstruction, Wikipedia
- Iterative reconstruction, Wikipedia
- Magnetic resonance imaging (introduction, History), Wikipedia
- History of magnetic resonance imaging, Wikipedia
- PET–CT (introduction, History), Wikipedia
- Convolutional neural network, Wikipedia
- Radiomics, Wikipedia
- Nuclear medicine, Wikipedia
The frameworks of Medical Imaging
Each framework is a way of working the field has used: when it held, and the question it answers.
- 1895-Present
Projectional radiography
How can X-ray shadows reveal bones and lesions inside the body?
- 1896-Present
- 1950-Present
- 1950-Present
Nuclear medicine
How can a radioactive tracer map where physiology happens?
- 1970-Present
Image Reconstruction and Computational Imaging
How can indirect measurements be turned into a faithful image?
- 1972-Present
- 1973-Present
Magnetic Resonance Imaging
How can magnetic resonance reveal soft-tissue contrast without radiation?
- 1980-Present
Digital Radiography and Imaging Informatics
How can radiological images become numbers that can be stored and processed?
- 1985-Present
Computer-aided diagnosis
Can a computer find lesions that a radiologist might miss?
- 1995-Present
Molecular Imaging
How can the molecules and processes of disease be imaged in a living patient?
- 2000-Present
Hybrid and Multimodality Imaging
How can different imaging signals be aligned into one dataset?
- 2010-Present
Radiomics
Can quantitative image features predict diagnosis and outcome?
- 2012-Present
Deep Learning in Medical Imaging
Can networks learn diagnostic features directly 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.
Fluoroscopy grew out of Projectional radiography
Fluoroscopy grew directly out of Röntgen's discovery of X-rays: within months of the 1895 announcement, investigators placed a fluorescent screen in the beam so the shadow image could be watched in real time rather than recorded on film. It kept radiography's differential-attenuation principle but added live observation.
Carried by Wilhelm Röntgen; Thomas Edison.
CT scan influenced by Projectional radiography
CT took the physical quantity of projectional radiography—X-ray attenuation along a line through the body—and measured it along hundreds of paths at many angles instead of collapsing it into one shadow. Hounsfield's scanner is unthinkable without the attenuation physics and clinical X-ray sources of radiography.
Carried by Godfrey Hounsfield; Allan Cormack.
Digital Radiography and Imaging Informatics grew out of Projectional radiography
Digital radiography is projectional radiography with the film emulsion replaced by electronic detectors—photostimulable phosphor plates and then flat-panel arrays—so the same attenuation shadow is captured as numbers. The clinical practice and exposure geometry passed over intact; only the recording medium changed.
Molecular Imaging grew out of Nuclear medicine
Molecular imaging generalises nuclear medicine's emission principle from imaging where a tracer accumulates to imaging specific molecules, receptors and gene activity. The radiopharmaceutical-and-detector logic of the gamma camera and PET is carried forward and aimed at molecular targets.
Hybrid and Multimodality Imaging influenced by Nuclear medicine
PET/CT fused the functional emission signal of nuclear medicine with CT's anatomical map in one scanner and one session. The hybrid design exists because PET alone lacks anatomical landmarks, so nuclear medicine supplied the functional half of the combination.
Magnetic Resonance Imaging influenced by Image Reconstruction and Computational Imaging
Lauterbur's 1973 MRI method is a reconstruction problem: magnetic field gradients encode spatial position into the NMR signal, and the image is recovered from those encoded measurements by projection reconstruction and later Fourier methods. MRI borrowed the reconstruction toolkit that CT had just made central.
Carried by Paul Lauterbur; Peter Mansfield.
Hybrid and Multimodality Imaging influenced by CT scan
PET/CT and PET/MRI use CT (and MRI) as the anatomical reference onto which the functional image is registered and overlaid. CT's fast, high-resolution anatomical map is what made combined functional-anatomical scanning clinically practical.
Computer-aided diagnosis influenced by Digital Radiography and Imaging Informatics
Computer-aided detection could only exist once images were digital arrays: CAD algorithms for mammographic microcalcifications and chest nodules operated on digitised or direct digital radiographs. The informatics turn made the pixels available to software, which is the precondition for automated lesion search.
Deep Learning in Medical Imaging influenced by Digital Radiography and Imaging Informatics
Deep learning in medical imaging depends on large labelled digital image archives, which PACS and digital radiography created. The conversion of radiology to numbers and networks supplied the training data and the pixel pipelines that convolutional networks consume.
Deep Learning in Medical Imaging replaced Computer-aided diagnosis
Deep learning replaced handcrafted-feature CAD as the working method of automated image analysis: convolutional networks learn their own features from labelled images instead of engineers specifying nodule roundness or edge gradients. The 2012 ImageNet result and subsequent medical applications defined themselves against the older CAD approach and its false-positive problems.
Carried by Alex Krizhevsky; ImageNet Classification with Deep Convolutional Neural Networks (2012).
Digital Radiography and Imaging Informatics influenced by CT scan
CT established that radiological images could be arrays of numbers processed by computer, and the PACS and informatics infrastructure built for CT was extended to projection radiography. The digital-image workflow that CT required became the model for digital radiography and hospital image networks.
Hybrid and Multimodality Imaging influenced by Magnetic Resonance Imaging
PET/MRI followed PET/CT by pairing functional emission imaging with MRI's soft-tissue contrast and lack of ionising radiation. MRI provided the anatomical and tissue-characterisation half of the hybrid scanner.
Radiomics influenced by Computer-aided diagnosis
Radiomics extends the CAD premise—that quantitative image features carry diagnostic information—from a few lesion descriptors to hundreds of extracted texture and shape features correlated with outcomes. The feature-extraction and classifier pipeline of computer-aided diagnosis is the direct ancestor of the radiomics workflow.
Hybrid and Multimodality Imaging influenced by Molecular Imaging
Hybrid scanners were built to localise molecular signals anatomically: the molecular imaging ambition of seeing receptors and processes required an anatomical map to say where in the body the signal arose. Multimodality imaging is the practical answer to that localisation problem.
Deep Learning in Medical Imaging influenced by Radiomics
Deep learning took over radiomics by replacing handcrafted texture features with features learned directly from images, and radiomics datasets and outcome-labelling practices fed the training of these networks. The quantitative-feature tradition supplied both the data culture and the benchmark tasks.
Hybrid and Multimodality Imaging influenced by Medical ultrasound
Ultrasound was combined with other modalities—most notably ultrasound with CT or MRI for registration and fusion-guided procedures—as part of the multimodality design principle. The link is real but minor compared with the PET/CT and PET/MRI combinations.
Beyond the field
Where Medical Imaging's ideas came from in other fields, and what they led to there: the strongest 12 of 84 links.
Projectional radiography influenced Oral and maxillofacial radiology in Dentistry
Projectional radiography supplied the intraoral and panoramic techniques that oral and maxillofacial radiology uses. The specialty is a direct application of that imaging method.
Fluoroscopy grew out of Radiography in Imaging Physics
Fluoroscopy grew directly out of radiography within a year of Röntgen's discovery: instead of a photographic plate, a fluorescent screen made the X-ray shadow visible in real time. It is a branch of the same radiographic technique.
Carried by Wilhelm Conrad Röntgen; Thomas Edison.
Medical ultrasound influenced POCUS in Critical Care Medicine
Medical ultrasound provided the imaging technology and transducer methods that clinicians later brought to the bedside as point-of-care examination.
Medical ultrasound influenced Targeted Ultrasound Molecular Imaging in Molecular Imaging
Medical ultrasound supplied the pulse-echo instrumentation, transducer technology and contrast-agent methods that targeted ultrasound molecular imaging adapted to detect bound microbubbles. The molecular variant is an extension of diagnostic ultrasound technique.
Medical ultrasound influenced by Ultrasound in Acoustics
Medical ultrasound applies the physics and instrumentation of acoustical ultrasound, developed through sonar and ultrasonic flaw detection, to tissue imaging. The pulse-echo method and transducer technology passed directly from acoustics into diagnostic medicine.
Carried by Paul Langevin; Floyd Firestone.
Medical ultrasound led to POCUS in Emergency & Critical Care
POCUS is a branch of medical ultrasound, taking the physics and instrumentation of diagnostic ultrasound and moving it to the bedside for immediate clinician-performed assessment. The technique descends directly from hospital ultrasound practice.
Medical ultrasound influenced Emergency ultrasound in Emergency Medicine
Emergency ultrasound is a direct application of medical ultrasound, taking its transducers, imaging modes and safety principles and adapting them to bedside emergency diagnosis.
Nuclear medicine led to Positron Emission Tomography in Neuroimaging
PET is a branch of nuclear medicine, using injected radiotracers for diagnosis; its makers came from nuclear medicine and adapted its tracer methods to positron emitters.
Carried by Michael Phelps; Ter-Pogossian.
Image Reconstruction and Computational Imaging led to Computational and Model-Based Imaging in Imaging Physics
Image reconstruction and computational imaging supplied the forward models, iterative algorithms and regularization that computational and model-based imaging built on after 2000.
Image Reconstruction and Computational Imaging influenced Magnetic Resonance Imaging in Neuroimaging
MRI reconstruction uses the Fourier and back-projection reconstruction machinery developed in computed tomography and computational imaging, adapted to spatial encoding by magnetic field gradients.
Carried by Peter Mansfield.
CT scan influenced by Tomographic Reconstruction in Inverse Problems
CT reconstruction rests on the mathematical tomographic reconstruction developed in inverse problems, above all the convolution-backprojection algorithm derived from the Radon transform. Hounsfield's scanner was built on that mathematics.
Carried by Johann Radon; Godfrey Hounsfield; Allan Cormack.
Magnetic Resonance Imaging led to Diffusion-weighted magnetic resonance imaging in Quantitative 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.

Medical Imaging is a planet in Nousfarer
Its city has a tower for each of these 13 frameworks. Land, walk in, and climb one concept at a time.
Land on Medical ImagingRelated fields
Fields its ideas are linked to
- Molecular Imaging 14 links
- Quantitative Imaging 11 links
- Imaging Physics 9 links
- Imaging Informatics 7 links
- Medical Image Analysis 5 links
- Neuroimaging 5 links
- Urology 4 links
- Biomedical Engineering 3 links
- Computer Vision 1 link
- Deep Learning 1 link
- Inverse Problems 1 link
- Art Connoisseurship 1 link
More in Medical Imaging
- Diagnostic Radiology
- Image Reconstruction
- Imaging Informatics
- Imaging Physics
- Interventional Radiology
- Medical Image Analysis
- Molecular Imaging
- Nuclear Medicine
- Quantitative Imaging
See where Medical Imaging sits on the Map of Human Knowledge