Ring Artifacts in Tomographic Imaging

Tomographic imaging, such as Computed Tomography (CT), has revolutionized medicine and science by allowing us to see inside objects non-invasively, creating detailed cross-sectional images slice by slice. While this technology provides incredible insight, the final images are sometimes marred by imperfections known as artifacts. Among the most common and recognizable are ring artifacts—faint, circular patterns that can obscure details and compromise quantitative accuracy. These "ghosts in the machine" raise critical questions: Are they just random noise, or do they signify a deeper problem with the scanner?

This article addresses the origin and implications of ring artifacts, peeling back the layers of the imaging process to reveal their cause. By understanding these imperfections, we not only learn how to eliminate them but also gain a powerful diagnostic tool for assessing the health of the imaging system itself. The following chapters will guide you through this process. First, in "Principles and Mechanisms," we will explore the fundamental physics and geometry of CT scanning, showing how a single faulty detector element inevitably leads to a circular artifact. Then, in "Applications and Interdisciplinary Connections," we will see how this knowledge is applied, turning a visual flaw into a precise quantitative measure for diagnosing instruments across fields from medical physics to materials science.

To understand the ghost in the machine that is the ring artifact, we must first appreciate the beautiful process by which a Computed Tomography (CT) scanner sees the world. It doesn't take a single picture, like a regular camera. Instead, it builds up a picture, slice by slice, by taking a series of X-ray "shadows" from hundreds of different angles as the source and detector rotate around the object.

Imagine a single, thin slice of an object. At one particular angle, the X-ray source shines a fan of beams through the slice, and a detector array on the other side measures what gets through. This one-dimensional "shadow" is called a ​​projection​​. It's a map of how much the X-ray beam was attenuated along each path.

Now, what do we do with the hundreds of projections taken as the scanner rotates? We stack them up. Imagine a 2D plot where the horizontal axis represents the position along the detector array, and the vertical axis represents the angle of rotation. Each projection is a horizontal row in this plot. The resulting image is called a ​​sinogram​​. It might look like a jumble of wavy lines, but it contains all the information needed to reconstruct the 2D slice.

There's a beautiful mathematical duality at play here, governed by an idea called the Radon transform. To get a feel for it, consider what a single, tiny, dense point in the object looks like in the sinogram. As the scanner rotates around it, the point’s shadow moves back and forth across the detector. Plotted in the sinogram, this movement traces a perfect sine wave—which is where the name "sinogram" comes from! Conversely, a circular object in the image space, like the cross-section of a wire, traces out two bounding sine waves in the sinogram, corresponding to the X-ray paths that are just tangent to its sides. This elegant correspondence between shapes in the image and patterns in the sinogram is the key to understanding both how CT works and how it can fail.

Now, let's turn the question around. Instead of asking what the object looks like to a perfect scanner, let's ask what a perfect object looks like to an imperfect scanner.

The detector array is made of hundreds or thousands of individual detector elements. What if just one of them is miscalibrated? Perhaps its electronics have a slightly higher background noise (an ​​offset​​ error), or its response to X-rays is a little more or less enthusiastic than its neighbors (a ​​gain​​ error). Let's say this faulty element is at a fixed position, $s_0$, in the detector array.

As the scanner rotates, this single detector element consistently reports a slightly wrong value, regardless of the angle, $\theta$. What does this look like in the sinogram? For every angle (every row in our sinogram image), there's a tiny error at the exact same horizontal position, $s_0$. This creates a perfectly straight, vertical line of faulty data in the sinogram. This vertical stripe is the "smoking gun," the unmistakable fingerprint of a detector-related problem. The error isn't in the object being scanned; it's in the scanner's "retina."

So, our raw data, the sinogram, is tainted with a vertical stripe. How does the computer’s reconstruction algorithm, a process called ​​Filtered Backprojection​​, turn this into the circular artifact we see in the final image?

Think of backprojection as the reverse of taking a shadow. The algorithm takes each one-dimensional projection and "smears" it back across the empty image canvas from the direction it was originally taken. When all these smeared-back projections are added up, the true features of the object reinforce each other and a clear image emerges.

Now, what happens to our vertical stripe of error? We are essentially smearing back a single point of error from every single angle. Imagine a paintbrush held at a fixed distance $s_0$ from the center of a canvas, always pointed toward the center. If you rotate the canvas underneath the brush, what do you get? A perfect circle.

This is precisely what the backprojection algorithm does. The line of error at position $s_0$ in the sinogram is mathematically transformed into a circle of radius $r = |s_0|$ in the reconstructed image. This is the ring artifact. Its origin is purely geometric. The radius of the ring tells you exactly which detector element is the culprit!

The "color" of the ring—whether it's dark or bright—tells you the nature of the fault. The scanner's computer calculates attenuation by taking a logarithm of the ratio of incoming to detected X-rays. Because of the properties of the logarithm, a multiplicative gain error in the detector becomes an additive error in the sinogram data. If a detector is over-responsive (gain $g_k > 1$), it registers more X-rays, making the object seem less attenuating along that path. This results in a darker, or ​​hypodense​​, ring. Conversely, an under-responsive detector ($g_k 1$) makes the object seem more attenuating, creating a brighter, or ​​hyperdense​​, ring in the final image.

The two essential ingredients for a classic ring artifact are therefore: (1) a detector-specific error, and (2) that error must appear at the same radial position, $s$, for all projection angles, $\theta$.

This simple rule reveals a fascinating and powerful insight into scanner design. In a common ​​3rd generation​​ scanner, the X-ray source and the detector arc rotate together as a single unit. A faulty detector element always maintains the same geometric relationship to the center of rotation. Its error therefore maps to a constant $s$ value for all angles, producing a perfect vertical stripe in the sinogram. This makes 3rd generation systems inherently susceptible to ring artifacts.

Contrast this with a ​​4th generation​​ scanner, where a full circle of detectors remains stationary while the source rotates inside it. Now, a single faulty detector is struck by X-rays from a continuously changing angle. Its error no longer maps to a constant $s$ value. Instead, it traces a sine wave through the sinogram! When backprojected, this distributed error doesn't create a sharp ring but is smeared into a much less conspicuous haze or a set of faint, crossing streaks. The same fundamental detector fault produces a dramatically different result, all because of a change in geometry.

This principle is stunningly general. The source of the error doesn't have to be a completely "dead" detector. Any subtle, uncorrected effect that creates a stable, detector-dependent bias can paint a ring. This includes:

• Imperfect ​​flat-field calibration​​, which is meant to correct for gain variations and non-uniformities in the X-ray beam itself, such as the ​​anode heel effect​​ or ​​off-focal radiation​​.
• A residual error in the ​​dark-field calibration​​, which measures the detector's electronic noise when the X-ray beam is off.
• Even a tiny, slow drift in a detector's gain during the few seconds of a scan can have its average effect manifest as a faint ring.

Because the cause is a detector-specific error, the appearance of rings is largely independent of the object being scanned. This makes them easy to distinguish from other common artifacts like ​​beam hardening​​, which creates "cupping" or dark streaks that depend heavily on the object's size and material composition.

By understanding the origin of the ring artifact, we not only learn how to diagnose it and correct it—through meticulous calibration procedures and clever software that can identify and interpolate over bad detector data—but we also gain a much deeper appreciation for the beautiful and sometimes fragile interplay of physics, geometry, and computation that makes modern imaging possible. It turns out that sometimes, the imperfections are the most instructive teachers of all. In a final ironic twist, some imperfections can even cancel others out; a slight wobble, or ​​rotational jitter​​, in the scanner's gantry can tangentially blur the rings, sometimes just enough to make them disappear from view.

In our last discussion, we peered behind the curtain of tomographic imaging and saw how the elegant mathematics of reconstruction can sometimes be haunted by ghosts—the faint, circular apparitions we call ring artifacts. We learned that these are not supernatural phenomena, but rather the direct consequence of imperfections in our detectors, a constant error in a single detector element that, when smeared across a full rotation by the backprojection algorithm, draws a perfect circle in the final image.

Now, you might be thinking of these artifacts as mere nuisances, flaws to be eliminated. And you are right, in a way. But in science, a flaw, when understood, often becomes a tool. An unexpected result is not a failure but a new discovery waiting to be interpreted. In this chapter, we will embark on a journey to see how these ghostly rings are not just problems to be solved, but windows into the heart of our imaging machines, guiding us in fields as diverse as dentistry, materials science, and nuclear medicine.

Imagine you are a medical physicist responsible for a multi-million dollar CT scanner. One morning, the radiologists report seeing faint rings in their images. Is the machine broken? Where do you even begin to look? The ring artifact itself is your first and most important clue.

The very presence of a ring tells you the problem likely lies not with the patient or the software, but with the hardware—specifically, the detector array. But we can do much better than that. The geometry of the ring is not arbitrary; it contains precise information. The radius of a ring artifact is a direct geometric fingerprint of the faulty detector element. By knowing the scanner's geometry—the distance from the X-ray source to the center of rotation ($R$) and to the detector ($D$)—we can work backward from the radius of the ring in our image to calculate the exact physical location of the misbehaving detector pixel on its panel. Isn't that marvelous? The ghost in the image points a finger directly at its source in the machine. A physician diagnoses a patient; a physicist uses an artifact to diagnose the instrument.

To be a true detective, however, we must learn to read the native language of the scanner, a language spoken not in pixels and images, but in sinograms. The sinogram is the raw data, a collection of all the attenuation measurements before the fancy reconstruction mathematics is applied. Looking at the sinogram is like a cardiologist reading an EKG rather than just observing the patient's symptoms. Different problems create wildly different patterns. A small, dense object, like a metal pin, traces a bright, elegant sine wave. A patient's sudden cough or movement shatters these sine waves into broken, discontinuous pieces. But a ring artifact? It announces itself with stark simplicity: a perfectly straight, vertical line. This unwavering line signifies a detector element whose error is constant, stubbornly independent of the rotation angle. By learning to distinguish these signatures—the vertical stripe of a ring, the jagged tear of motion, the sinusoidal track of metal—an engineer can diagnose the health of a scanner with astonishing precision, often before the artifacts even become obvious in the final image.

Identifying the source of the problem is one thing; understanding its severity is another. A faint ring might be a cosmetic issue, but a pronounced one represents a significant error in the scanner's primary job: measuring the physical properties of the material it is imaging.

Computed Tomography doesn't just produce pretty pictures; it produces quantitative maps of a physical quantity, the linear attenuation coefficient $\mu$. In the medical world, we scale these values into Hounsfield Units (HU), where water is defined as $0$ HU and air is $-1000$ HU. A physician may make a diagnosis based on whether a tissue's value is $30$ HU or $50$ HU. So, what does a ring artifact mean in these terms?

Let's imagine a small, uncorrected drift in a detector's gain—say, it reports being hit by just $0.5\%$ more light than it actually received. It seems tiny. But through the machinery of logarithms and reconstruction, this small instrumental error can manifest as a noticeable bias in the final image. We can calculate that this tiny gain drift can create a dark ring that is several Hounsfield Units different from the surrounding tissue. This is no longer just a ghost; it's a lie written into the data, a quantitative falsehood that could, in principle, affect clinical decisions.

This principle extends beyond the clinic. In materials science, researchers use techniques like synchrotron-based micro-tomography to create ultra-high-resolution 3D images of things like metal alloys or composite structures. Here, the goal is to measure subtle variations in density or composition. If the detector elements have, on average, a random gain variation of, say, $2\%$, then the reconstructed image will be contaminated with a flurry of faint rings, and the root-mean-square (rms) intensity of these artifacts will be, remarkably, about $2\%$ of the background signal itself. The statistical uncertainty in the instrument's components propagates directly into a measurable uncertainty in the final scientific result. The quality of the image is tethered directly to the quality of the detector.

One of the most beautiful things in physics is the discovery of a universal principle, an idea that explains seemingly unrelated phenomena. The story of ring artifacts is one such case. We have spoken of CT scanners used in medicine and materials science, which work by transmitting X-rays through an object. But what about a completely different technology, like Positron Emission Tomography (PET)?

A PET scanner doesn't send radiation through the body; it detects gamma rays emitted from a radioactive tracer within the body. Its purpose is to map metabolic function, not just anatomical structure. It's a completely different physical process. And yet, PET images can also suffer from ring-like or star-like artifacts. Why? Because the fundamental situation is the same: a ring of detectors with non-uniform sensitivity that isn't perfectly corrected. In PET, the "imperfection" might not be electronic gain, but rather the physical gaps between detector blocks, or the tendency for a gamma ray to scatter from one crystal to another. These effects cause some angles of detection to be more or less sensitive than others. If the reconstruction algorithm assumes all directions are created equal, this periodic variation in sensitivity around the detector ring will be misinterpreted as a feature in the object, creating a structured, circular artifact. The context changes, the physics changes, but the core mathematical story—uncompensated systematic variation in a circular acquisition—remains the same.

If understanding artifacts is the science, then eliminating them is the engineering. The most fundamental method of correction is beautifully simple in concept. It's called ​​flat-field correction​​. Before you scan the object, you perform two calibration scans: one with the X-ray beam on but nothing in the scanner (an "air" or "flat" field), and one with the beam off (a "dark" field). The dark field measures the electronic offset of each detector pixel, and the air scan measures its unique response to a uniform X-ray beam, capturing its individual gain. By subtracting the dark field from all measurements and then dividing the object scan by the air scan, we can, in a single mathematical stroke, cancel out the effects of both offset and gain for every single pixel. This elegant procedure is the first line of defense against ring artifacts and a foundational step for every tomographic reconstruction, from the simplest algorithms to the most advanced Deep Learning networks.

Of course, the world is rarely so simple. Detectors can drift over time, so a calibration scan from this morning might not be perfect for a scan this afternoon. This is why advanced systems may require temporally matched calibrations or employ clever software filters that hunt for and suppress the tell-tale vertical stripes in the sinogram before they ever become rings in the image.

We can even automate the search for these ghosts. By transforming the image into the frequency domain using a Fourier transform, we move into a world where spatial location is replaced by spatial frequency, or texture. A ring artifact, which is spread out across the image, becomes wonderfully concentrated in the frequency domain. It appears as a sharp, bright circle against the more diffuse background of the image's "true" content. This makes it incredibly easy for an algorithm to detect, quantify, and even filter out. It is a classic strategy in physics: if a problem is hard to see in one representation, change your point of view, and it might become simple.

From a diagnostic clue in a hospital to a source of quantitative error in a materials lab, from a problem in X-ray transmission to one in gamma ray emission, the ring artifact teaches us a profound lesson. It reminds us that our instruments are not perfect, and that the path from measurement to image is a delicate dance of physics and mathematics. By studying these ghosts in the machine, we learn not only how to banish them, but also how to build better instruments and push the boundaries of what we can see.