Motion Artifact

In the precise world of scientific measurement, a fundamental assumption is often made: the subject of study remains perfectly still. However, when dealing with living systems—from a patient in an MRI scanner to a person jogging with a smartwatch—this assumption is inevitably broken. The result is the motion artifact, a phantom in the data that can obscure truth, mimic disease, and challenge the very foundation of our measurements. These artifacts are not simply random noise; they are structured distortions born from the predictable interaction between motion and measurement physics. Understanding them is critical for anyone who collects or interprets data from the real, dynamic world.

This article confronts the challenge of motion artifacts head-on. The first section, ​​Principles and Mechanisms​​, will dissect the fundamental physics behind how movement corrupts data, exploring a 'rogues' gallery' of artifacts from simple blurs to complex, fictitious signals across various technologies. We will examine their tell-tale signatures and uncover the profound danger they pose by masquerading as genuine biological findings. Following this, the section on ​​Applications and Interdisciplinary Connections​​ will climb a ladder of solutions—from procedural controls to sophisticated signal processing and revolutionary measurement techniques—revealing how the fight against artifacts has spurred innovation across medicine, engineering, and neuroscience. Our journey begins by exploring the core of the problem: the elegant, yet flawed, assumption of stationarity.

To understand a motion artifact is to appreciate a fascinating collision between the elegant, ordered world of physics and the messy, unpredictable reality of living things. At their heart, most of our powerful scientific instruments—from towering MRI scanners to tiny wrist-worn sensors—are built on a simple, almost naive, act of faith: that the object they are measuring will hold perfectly still. A motion artifact is the spectacular and often bewildering result of that faith being broken. It is the ghost in the machine, a phantom born from the betrayal of this fundamental assumption of ​​stationarity​​.

Imagine you are taking a photograph in a dimly lit room. To capture enough light, you must use a long exposure, perhaps a full second. If your subject—a smiling friend—remains perfectly still, you get a sharp, clear portrait. But if your friend sneezes, laughs, or even just wobbles slightly during that one second, the camera, in its unwavering gaze, dutifully records light from every position your friend's face occupied. The result? A ghostly blur, a smear of light where a crisp edge should be.

This is the essence of a motion artifact. A computed tomography (CT) scanner, for example, builds its intricate cross-sectional images by combining hundreds of X-ray projections taken from different angles around the body. The reconstruction algorithm is a mathematical masterpiece that assumes every single one of those projections is looking at the exact same, stationary object. If the patient breathes, the diaphragm moves, and a projection taken at the beginning of the breath sees the liver in one place, while a projection taken moments later sees it in another. The algorithm, unaware of this betrayal, tries to reconcile these inconsistent views, resulting in characteristic streaks, blurs, or double images. The data no longer represents a single, consistent reality, and the final image is a testament to this contradiction.

While the underlying cause is always a violation of stationarity, the ways in which motion artifacts manifest are as varied and creative as the measurement techniques themselves. They are not one single entity but a veritable rogues' gallery of distortions, each with its own character shaped by the physics of the device and the nature of the movement.

The Usual Suspects: Blurs, Streaks, and Ghosts

The most familiar artifact is, of course, the blur. In signal processing terms, this kind of motion can often be modeled as a ​​convolution​​—the mathematical equivalent of smearing the ideal, sharp image with a blurring function. This process acts as a ​​low-pass filter​​, selectively wiping out the high spatial frequencies that define fine details and sharp edges. This is why motion can make a textured, complex tumor appear artificially smooth, a change that can have serious consequences for quantitative analysis like radiomics.

When the motion is periodic, like the steady rhythm of a heartbeat or respiration, the artifacts can become more structured. Instead of a simple blur, we might see distinct "ghost" replicas of the moving anatomy, or sharp streaks that seem to radiate from high-contrast objects. These are the regular, repeating patterns of data inconsistency being systematically woven into the final image by the reconstruction algorithm. Some artifacts, like the "windmill" pattern seen in helical CT scans, are even more complex, arising from the interplay between periodic physiological motion and the specific corkscrew-like path the scanner takes through the body.

Creative Corruptions: When Motion Does the Unexpected

Sometimes, motion does more than just blur or streak. It can interact with the measurement process in much more insidious ways, creating signals that are entirely fictitious.

A striking example occurs in Magnetic Resonance Imaging (MRI). To select a specific slice of the body to image, the MRI scanner applies a magnetic field gradient and a precisely tuned radiofrequency (RF) pulse. This process takes a few milliseconds. If the patient moves along the direction of the slice selection during this brief window of time, a part of the body that was initially outside the target slice can move into the region of excitation, while a part that was inside can move out. The scanner, none the wiser, proceeds to create an image of what it thinks is the prescribed slice, but the content of that image is actually from a neighboring anatomical location. This is ​​motion-induced slice misregistration​​—not just a blurred image, but an image of the wrong thing entirely.

Another form of creative corruption is the "flash artifact" in Doppler ultrasound. This technique measures blood flow by detecting the frequency shift in sound waves bouncing off moving red blood cells. To isolate the weak signals from blood, the machine uses a "wall filter" to ignore the strong, low-frequency echoes from slowly moving tissue. However, if the operator accidentally jostles the ultrasound probe, the entire mass of tissue in the image can appear to move with a sudden, high velocity. This generates a powerful Doppler signal with a frequency high enough to pass through the wall filter. It can even be so fast that it exceeds the system's measurement limit (the Nyquist frequency) and experiences ​​aliasing​​, appearing as flow in the opposite direction. The result is a brilliant, overwhelming flash of color on the screen that completely swamps the true, delicate signal of the blood flow it was meant to display.

In the world of wearable sensors, motion can even engage in a kind of duet with the signal of interest. The signal from a Photoplethysmography (PPG) sensor on a smartwatch, which measures blood volume changes in the wrist, is essentially a carrier wave at the frequency of the heart rate, $f_H$. When you walk or run, the periodic motion of your arm swing, at a frequency $f_m$, doesn't just add noise—it can physically modulate the contact of the sensor and the pressure on the underlying tissue. This results in amplitude modulation, a multiplicative effect. In the frequency spectrum, this doesn't just create a peak at the motion frequency $f_m$; it creates new ​​sidebands​​ flanking the cardiac peak, at frequencies of $f_H + f_m$ and $f_H - f_m$. The artifact is not merely added to the signal; it is woven into its very fabric.

If we are to have any hope of dealing with these phantoms, we must first learn to recognize them. Fortunately, motion artifacts have characteristic "fingerprints" or signatures in the data, which distinguish them from both true biological signals and other sources of noise.

The most important signature is in the frequency domain. Most bulk physiological movements—breathing, heartbeats, limb motion, head turning—are slow processes. Their energy is overwhelmingly concentrated in the low-frequency part of the spectrum, typically below 20 Hz. In contrast, many true physiological signals, like the neural firings that produce an electromyography (EMG) signal, are rich in high frequencies. This spectral separation is a gift. It means we can often design a high-pass filter that mercilessly cuts out the low-frequency motion artifact while preserving the high-frequency signal of interest. This also distinguishes motion artifact from electronic thermal noise, which is typically "white"—its power is spread evenly across all frequencies—whereas motion artifact is "colored," with a power spectrum that often follows a $1/f$ pattern, rising dramatically at the lowest frequencies.

In the time domain, motion artifacts are characteristically ​​non-stationary​​. Unlike the persistent hiss of thermal noise, motion artifacts come and go. They appear as sudden bursts or drifts in the signal that coincide with a physical event. A classic example from neurophysiology is the "electrode pop," which occurs when an EEG electrode momentarily loses good contact with the scalp. This creates an abrupt, step-like jump in the recorded voltage. When this step function passes through the AC-coupled amplifier—which is itself a high-pass filter—it is transformed into a sharp peak followed by a long, slow exponential decay back to baseline. The duration of this disruptive tail is determined entirely by the filter's time constant. A lower cutoff frequency means a longer time constant, and thus a longer period of data corruption from a single, instantaneous event. This principle of linking transient physical events to specific signal morphologies is the foundation of artifact identification in complex recordings like overnight sleep studies, which are rife with signatures from eye blinks, muscle tension, swallowing, and respiratory movements.

Perhaps the greatest danger of motion artifacts is not that they make our data look ugly, but that they can be mistaken for genuine scientific findings. They can be phantoms that masquerade as biology.

Nowhere is this clearer than in the field of resting-state functional MRI (fMRI), which seeks to map the brain's intrinsic communication networks by measuring correlations in spontaneous brain activity. One of the most prominent networks is the Default Mode Network (DMN), characterized by strong correlations between brain regions that are very far apart. A pernicious effect known as ​​distance-dependent motion artifact​​ poses a direct threat to the study of such networks.

Here is how it works. When a person moves their head in the scanner, it adds a noise component, $\epsilon(t)$, to the true neural signal, $s(t)$, at every location. For two brain regions that are close together, the motion-induced noise is very similar. This shared noise artificially inflates the measured correlation between them. However, for two regions that are far apart—like the nodes of the DMN—the motion-induced noise is largely uncorrelated. This adds random variance to both signals without adding to their shared covariance. The Pearson correlation is calculated as covariance divided by the product of standard deviations. By adding uncorrelated noise, motion increases the denominator of this fraction without increasing the numerator, thus systematically and artificially reducing the measured correlation.

The stunning consequence is this: a person who moves more during a scan will appear to have weaker long-range brain connectivity and stronger short-range connectivity. A researcher might erroneously conclude that a disease group has impaired DMN integrity, when in fact that group simply found it harder to lie still in the scanner. The artifact has perfectly mimicked a profound neurological finding.

This theme echoes in other quantitative fields. An apparent decrease in a tumor's textural complexity, a feature often associated with a better prognosis in radiomics, might be nothing more than the low-pass filtering effect of patient motion during the CT scan. The principles and mechanisms of motion artifacts are therefore not just a technical curiosity for engineers. They are a critical, foundational topic for any scientist or clinician who hopes to interpret data from the real, moving, and beautifully messy world. To see the artifact is to begin to see the truth.

In our journey so far, we have explored the fundamental principles of motion artifacts, viewing them not as mere errors, but as physical phenomena governed by the same laws that produce the signals we wish to measure. We’ve seen how an object's unexpected movement during the delicate process of observation can distort our picture of reality. But this is not where the story ends. In fact, it is where the real adventure begins. For in science and engineering, understanding a problem is the first step toward solving it, and the challenge posed by motion artifacts has sparked a remarkable display of human ingenuity across a breathtaking range of disciplines.

The confrontation with this "unwanted dance" of moving matter has led to a hierarchy of brilliant strategies. These strategies are not just a collection of ad-hoc tricks; they form a beautiful intellectual ladder, starting from the most direct physical interventions and climbing toward elegant mathematical and computational abstractions. Let us climb this ladder together and see how the battle against motion artifacts has not only cleaned up our data but has also deepened our understanding of the world and forged powerful connections between seemingly disparate fields.

The most straightforward way to deal with an unwanted dance is to gently ask the dancer to sit still. In the world of measurement, this translates to designing experiments and procedures that minimize motion at its very source. This approach is not a matter of brute force, but of subtle, intelligent control, blending physics with physiology.

Consider the challenge of assessing the health of a woman's reproductive system using ultrasound. In a procedure like hysterosalpingo-contrast sonography, a clinician introduces a contrast agent to visualize the uterine cavity and fallopian tubes. The clarity of the resulting image is paramount. But the human body is a symphony of motion—the gentle rhythm of breathing, the involuntary contractions of muscles, the peristaltic motion of the nearby bowel. Each of these movements can blur the image, making a diagnosis difficult or impossible.

The solution is a masterclass in interdisciplinary thinking. To reduce motion from uterine cramps, a nonsteroidal anti-inflammatory drug (NSAID) might be given beforehand, targeting the specific biochemical pathways that cause the contractions. To prevent bowel gas from scattering the ultrasound waves and creating its own form of "noise," a simple dietary or preparatory change can be advised. The patient is instructed to empty her bladder, which, for a transvaginal ultrasound, brings the uterus closer to the transducer, shortening the acoustic path and improving resolution. She is even coached in slow, deliberate breathing. This is not just a checklist; it is a clinical protocol born from a deep understanding of physics and human physiology, all aimed at achieving a moment of diagnostic stillness.

This principle extends far beyond the clinic. In a neurophysiology laboratory studying the vestibular system—our body's internal gyroscope—scientists measure tiny muscle potentials below the eye called oVEMPs. These signals are minuscule, and they are easily swamped by the electrical noise generated by the eye's own movement. To get a clean recording, the subject must gaze upwards, but this very act can introduce unwanted eye jitters. The solution is again one of careful, deliberate control: the subject's head is stabilized, and they are asked to fixate on a stationary target placed at a specific distance and angle—far enough to reduce the strain of focusing, and angled just so, to activate the target muscle without causing excessive fatigue or instability. The experiment itself is choreographed to create a pocket of tranquility in which the whisper of the neural signal can be heard.

Sometimes, we cannot stop the motion. A fetus will squirm in the womb; a patient will inevitably move. In these cases, the scientist must become a detective, armed with skepticism and a keen eye for distinguishing the real signal from the impostor.

Imagine a physician peering at an ultrasound screen, trying to find the first signs of life in a six-week-old embryo. A subtle flicker appears. Is it the nascent beat of a tiny heart, or is it an artifact, a phantom created by the slight tremor of the sonographer's hand or the transmitted pulsation from the mother's own arteries? The stakes could not be higher. Here, the solution is not a single action but a process of rigorous verification. The detective-physician first immobilizes the probe. If the flicker vanishes, it was an artifact. If it persists, the investigation continues. Using a different mode of ultrasound, M-mode, which plots motion over time, a rhythmic pattern can be documented. The rate is measured. Is it around $120$ beats per minute, typical for a fetus of this age, or is it closer to the mother's pulse of $80$? By comparing, contrasting, and using multiple forms of evidence, the physician can confidently distinguish the signal of life from the noise of the living.

This same detective work occurs in the abstract realm of computational neuroscience. When recording the simultaneous activity of hundreds of neurons, a major challenge is that movement of the animal can cause the brain tissue itself to shift slightly relative to the electrodes or microscope. This creates a powerful, coordinated signal across many neurons that has nothing to do with neural computation, but is instead a pure motion artifact. This artifact can be so strong that it appears as the most prominent pattern in the data, a "principal component" that could easily be mistaken for a genuine, widespread neural dynamic.

Here, the neuroscientist's "clues" come from other sensors. An accelerometer attached to the subject provides an independent record of movement. By comparing the time course of the suspicious neural pattern to the accelerometer data, a strong correlation may be found. This "triangulation"—finding that a signal in the brain data looks, feels, and smells like the motion recorded by an external sensor—is the smoking gun. It allows the scientist to identify the artifactual component and digitally separate it from the underlying neural activity, ensuring that they are studying the brain's computations, not the jiggle of the apparatus.

When motion is unavoidable and its artifacts are mixed in with our data, we turn to the power of mathematics and computation. If we cannot create stillness in the physical world, we can create it in the digital one. The key insight is that often, the "signal" and the "noise" of motion sing in different keys.

A surface electromyography (EMG) recording, which measures the electrical activity of muscles, is a perfect example. The true EMG signal, representing the firing of motor units, has its energy concentrated in a specific frequency band, say from $20$ to $450$ Hertz ($Hz$). Motion artifacts, caused by the skin and electrode shifting, are typically slow, low-frequency events, mostly below $20\,Hz$. At the same time, electrical interference from power lines creates a sharp, high-frequency hum at exactly $50$ or $60\,Hz$. The raw recording is a cacophony of all three.

Signal processing provides a "digital sieve" to sort this out. A high-pass filter is designed to let all frequencies above $20\,Hz$ pass through, effectively cutting out the low-frequency rumble of the motion artifact. A low-pass filter lets all frequencies below $450\,Hz$ pass, removing higher-frequency electronic noise. Finally, a narrow "notch" filter is applied to specifically cut out the $50/60\,Hz$ power-line hum. What remains after passing through this gauntlet of filters is a clean signal, revealing the muscle's true activity.

But what if the artifact and the signal sing in the same key? This happens constantly in wearable sensors, like the photoplethysmogram (PPG) sensors in smartwatches that measure heart rate. The pulsatile change in blood volume from a heartbeat is a slow signal, often around $1-2\,Hz$. Unfortunately, the rhythm of walking or running also falls in this exact frequency range. A simple filter cannot tell them apart.

The solution is a wonderfully clever idea called Adaptive Noise Cancellation (ANC), which uses a second source of information. Your smartwatch contains not just a PPG sensor but also an accelerometer, which measures motion. The ANC algorithm operates on a simple principle: it uses the accelerometer signal as a template for the "noise." It continuously learns a filter that predicts the motion artifact component in the PPG signal based on the current accelerometer reading. It then subtracts this predicted artifact from the raw PPG signal in real-time. The result is a clean waveform of the heartbeat, as if the motion had never happened. It is the digital equivalent of noise-canceling headphones, which listen to the outside world with a microphone and subtract that sound from what you hear. This same adaptive philosophy is at the heart of more advanced methods like the Kalman filter, which dynamically adjusts its "belief" about the quality of a measurement based on how much it deviates from prediction—a large, unexpected deviation (an "innovation") during a period of known motion causes the filter to temporarily "distrust" the measurement, effectively down-weighting the artifact's influence.

The most elegant solutions transcend mere correction. They involve redesigning the very process of measurement to be inherently robust to motion. This represents the pinnacle of our intellectual ladder, where a deep understanding of physics allows us to invent entirely new ways of seeing.

Magnetic Resonance Imaging (MRI) provides a stunning example. An MRI scanner builds an image not all at once, but by patiently collecting data point by point in a mathematical space known as $k$-space, which is then converted into the final image via a Fourier transform. The process can take minutes. If the subject—say, a fetus in the womb—moves during this time, different parts of $k$-space will correspond to different positions of the head, and the resulting image will be a blurry, ghost-ridden mess.

For years, this made high-quality fetal MRI nearly impossible. The solution, which came under names like PROPELLER or BLADE, was a complete rethinking of how $k$-space is filled. Instead of filling it line by line in a simple grid, the scanner acquires the data in overlapping rectangular "blades" that are rotated around the center. The key is that the center of $k$-space, which contains the most fundamental information about the image's contrast and overall structure, is measured over and over again with each blade. This massive redundancy acts as a "self-navigator." By comparing the data from the center of each blade, the computer can calculate precisely how the fetus translated and rotated between the acquisition of one blade and the next. It can then digitally rotate and shift each blade's data back into a common reference frame before combining them to create the final image. It is a breathtaking fusion of physics, engineering, and computation that corrects for motion before the image is even formed.

The clinical impact of such an invention is profound. In a tragic condition like congenital high airway obstruction syndrome (CHAOS), a fetus's airway is blocked. Ultrasound can spot the secondary signs, like enlarged lungs, but it often struggles to pinpoint the exact location of the blockage due to its own physical limitations. MRI, armed with motion-correction techniques like PROPELLER, can provide crystal-clear images of the fluid-filled airway, allowing surgeons to map the obstruction with millimeter precision and plan a life-saving intervention. An innovation born from the abstract world of Fourier transforms and $k$-space directly enables a surgeon to save a newborn's life.

From the clinic to the laboratory, from wearable sensors to multi-million-dollar imaging systems, the challenge of motion artifacts has been a relentless and fruitful driver of innovation. The struggle has forced us to be better clinicians, more careful experimentalists, cleverer mathematicians, and more inventive physicists. It has revealed the deep unity of the scientific endeavor, where a principle from signal processing can save a life, and an observation in neuroscience can be explained by the physics of light and motion.

Ultimately, the presence of an artifact reminds us that every measurement is a story. Often, we only want to hear one story—the beat of the heart, the firing of the neuron. But the universe is telling countless stories at once. Motion artifacts are simply other stories intruding on our own. The ongoing quest to understand, identify, filter, and design around them is the quest to become better listeners—to learn how to isolate the single story we need to hear from the beautiful, complex, and ever-moving cacophony of the world. And as the need for rigorous quality control in fields like cancer diagnostics shows, learning to listen correctly is not just an academic exercise; it is a matter of profound human consequence.