Acoustic Neural Field Imaging (ANFI) is a hypothetical brain imaging modality that aims to noninvasively detect fast neural activity by using ultrasound to sense tiny mechanical vibrations in the brain. The idea is that when neurons fire action potentials, their membranes undergo slight mechanical movements and generate pressure waves in surrounding tissue. ANFI would use high-frequency ultrasound sensors arranged around the head to detect these tiny changes and translate them into maps of neural activity. This technology would show brain electrical activity, potentially achieving high temporal and spatial resolution without needing to use electrodes. In this report, we discuss what ANFI would measure, how the equipment might be designed, the nature of the data produced, key limitations and challenges, and the new research that could be made with this idea.
Acoustic Neural Field Imaging: A Hypothetical Ultrasound-Based Brain Imaging Method
Introduction
What Would It Measure?
ANFI would measure the tiny mechanical deformations that occur in neuronal membranes during action potentials. When a neuron fires, ion fluxes and voltage changes cause the membrane to move by a few nanometers. Early experiments showed that action potentials are accompanied by small diameter changes in invertebrate axons, typically around 0.3 to 5 nm. (Hill et al.,1977) used laser interferometry to detect about a 0.3 nm expansion in a crayfish giant axon during a spike, providing some of the earliest quantitative measurements of these movements. More recent interferometric imaging has confirmed similar nanometer scale displacements in mammalian cells (Ling et al., 2020). These movements arise from electromechanical coupling in the membrane, where changes in voltage alter membrane tension and shape.
The mechanical motion of the membrane produces a small pressure wave in the surrounding tissue. Models predict roughly 1 nm membrane displacement during each action potential, which generates an acoustic pulse that travels through intracellular and extracellular fluid. ANFI would detect these pressure changes using ultrasound sensors placed around the head.
Measuring these mechanical traces provides a direct window into neural firing, unlike fMRI, which measures slow blood-flow changes, or EEG, which reflects large-scale electrical fields. Because each action potential produces a brief mechanical “shake,” ANFI could potentially track neural activity at sub-millisecond timescales and with spatial detail beyond what surface electrical recordings allow. Earlier studies on axonal swelling and fast nerve expansion support the idea that mechanical signals consistently accompany electrical activity, which ANFI would take advantage of to map brain activity noninvasively.
System Design and Equipment Architecture
To capture brain wide acoustic signals from neural activity, ANFI would use specialized ultrasound instrumentation, likely in the form of a wearable transducer array. The device would resemble an ultrasound helmet containing many transducers positioned around the head, each capable of emitting focused pulses and receiving returning echoes. Recent advances in transcranial ultrasound make this design realistic. For example, a 256 element helmet shaped array operating at about 0.5 MHz has been developed for noninvasive neuromodulation and imaging in humans (Martin et al., 2025). This system uses piezoelectric transducers arranged in a semi ellipsoidal bowl filled with a water coupling medium to transmit sound through the skull. The head is stabilized with a face mask and frame to maintain alignment. ANFI would likely follow a similar layout, using a dense array of elements to capture acoustic signals from multiple angles.
The ultrasound array would operate in pulse echo mode to scan the brain’s volume. It would function similarly to ultrafast ultrasound systems that use multiple channels to send brief bursts of high frequency sound, typically between 250 and 700 kHz, through the skull. Frequency selection creates a trade off. Higher frequencies above 1 MHz provide better spatial resolution but are heavily attenuated by bone, while lower frequencies around 250 to 500 kHz penetrate the skull more effectively but give lower resolution. For transcranial applications, mid range frequencies offer the best balance between penetration and image clarity (O’Reilly and Hynynen, 2012; White et al., 2006). Systems operating near 500 to 600 kHz can achieve focal spots only a few millimeters wide in deep brain tissue. After each pulse, echoes return to the array and are converted into electrical signals. Small shifts in echo phase or frequency indicate subtle tissue movement that could reflect neural vibrations.
ANFI is designed as a fully non-invasive, wearable device, most likely in the form of a helmet or cap connected to external electronics. Current ultrasound helmets are bulky and require head stabilization, but future versions could use flexible arrays or modular panels that better conform to the scalp. The system may not be lightweight at first, but it could still be portable and usable at the bedside. It would require no implants, genetic modification, magnetic fields, or radioactive tracers, making it safer and more accessible than many existing brain imaging methods. A battery powered version may also be possible with further advances in efficiency.
What Would the Data Look Like?
The data produced by ANFI would be very much like a movie of the brain’s activity. Each frame of the movie is an ultrasound image (or slice) of the brain, where pixel intensities indicate blood flow strength or another activity-related signal. When neurons in a region become active, the subsequent increase in local blood volume causes a stronger ultrasound Doppler signal from that area (Rabut et al. 2021). Over the time course of a stimulus or task, one would see certain regions in the ultrasound images light up (become brighter in Doppler power) relative to baseline, indicating increased neural activity there. By acquiring frames in rapid succession, ANFI can capture the dynamics of these changes. For example, which brain areas activate first and how the activity spreads or fluctuates.
In technical terms, the raw data are often stored as power Doppler images. These are processed images computed from the backscattered echoes of the ultrasound pulses, after filtering out static tissue signals. What remains is primarily the signal from moving red blood cells, quantified as Doppler power that correlates with cerebral blood volume (Rabut et al. 2021). A series of such images over time can be analyzed to produce maps of brain activation. The spatial resolution of current ANFI (fUS) systems is around 100 × 100 × 300 μm for a 2D imaging plane at 15 MH (Rabut et al. 2021), which is fine enough to resolve small vascular units like cortical columns. If multiple 2D slices are acquired or a 2D matrix transducer is used, 3D volumetric data can be obtained. Recent demonstrations have shown that nearly whole-brain 3D imaging in small animals is possible by using matrix arrays (with many more ultrasound elements and channels) and scanning very fast. These 3D datasets can reveal functional activity throughout the brain volume, not just in a single slice.
It’s important to note that the temporal aspect of the data is limited by biology: the blood flow changes do not happen instantaneously when neurons fire. There is typically a delay of a few hundred milliseconds for blood flow to increase after neural activity, and the peak of the response might be ~1 second after a burst of neuron firing (Deffieux et al. 2018) ANFI captures this hemodynamic response. So while the imaging system might acquire frames at, say, 10 frames per second (100 ms per frame), the actual neural signal it’s measuring (blood volume changes) is smeared over a second or more by the neurovascular coupling process (Deffieux et al. 2018). Researchers often account for this by correlating the timing of stimuli or neural events with the slowly rising ultrasound signal. Despite the lag, the high frame rate of ultrasound is still beneficial because it improves sensitivity (by averaging many frames you get better signal-to-noise) and it can capture subtle transient events that slower frame rates would miss. In summary, ANFI data consists of high-resolution ultrasonic images or volumes over time, which are analyzed to extract when and where neural activity (inferred from blood flow or other acoustic signals) occurred.
Limitations
Despite its potential, ANFI faces several major limitations. One of the most significant is skull attenuation. The skull absorbs and scatters ultrasound, making it difficult to detect faint signals originating deep in the brain. Higher frequencies provide better spatial resolution but penetrate the skull poorly, forcing a trade-off between clarity and depth (O’Brien, 2007). Another challenge is the extremely small size of the neural mechanical signals. Nanometre-scale movements are close to the noise floor of typical ultrasound systems, and they occur in a living environment full of other mechanical activity, such as heartbeat, breathing, blood flow, and tiny head motions. Recent interferometric work emphasizes how careful one must be to separate actual neural deformations from various mechanical and optical artifacts (Ling et al., 2020). This means ANFI would require very sensitive sensors, low-noise electronics, and sophisticated analysis methods to distinguish neural signals from background noise. The engineering is also demanding: building and calibrating a helmet with hundreds of ultrasound channels, each with precise timing, and then processing all of that data in real time is a significant hardware and software challenge. On top of that, safety limits on ultrasound intensity must be respected to avoid heating or other bioeffects in brain tissue, which restricts how much power can be used to boost signal strength (O’Brien, 2007).
Applications
Despite these obstacles, ANFI could be very powerful if it becomes feasible. By targeting mechanical signatures tied directly to action potentials, it could provide a way to study brain activity with both wide coverage and high temporal resolution. For example, it could help researchers investigate how signals spread through distributed networks during perception, decision making, or memory encoding, in much finer temporal detail than fMRI allows. In clinical contexts, ANFI might help locate the onset zone of epileptic seizures more precisely than EEG, without requiring invasive electrodes. It might also support future brain computer interfaces by providing a noninvasive way to read out fast neural patterns that can be decoded into control signals. The concept shows how combining knowledge from electrophysiology, neuronal mechanics, and ultrasound physics could lead to a new type of brain imaging. While ANFI remains hypothetical and faces real physical and technical limits, it outlines a plausible future direction in which we do not just measure the brain’s electricity or blood flow, but also its subtle mechanical “voice” as neurons fire.
References
- Deffieux, T., Demene, C., Pernot, M., & Tanter, M. (2018). Functional ultrasound neuroimaging: a review of the preclinical and clinical state of the art. Current Opinion in Neurobiology, 50, 128–135. https://doi.org/10.1016/j.conb.2018.02.001
- Hill, B. C., Schubert, E. D., Nokes, M. A., & Michelson, R. P. (1977). Laser interferometer measurement of changes in crayfish axon diameter concurrent with action potential. Science (New York, N.Y.), 196 (4288), 426–428. https://doi.org/10.1126/science.850785
- Ling, T., Boyle, K. C., Zuckerman, V., Flores, T., Ramakrishnan, C., Deisseroth, K., & Palanker, D. (2020). High-speed interferometric imaging reveals dynamics of neuronal deformation during the action potential. Proceedings of the National Academy of Sciences, 117 (19), 10278–10285. https://doi.org/10.1073/pnas.1920039117
- Martin, E., Roberts, M., Grigoras, I. F., Wright, O., Nandi, T., Rieger, S. W., Campbell, J., den Boer, T., Cox, B. T., Stagg, C. J., & Treeby, B. E. (2025). Ultrasound system for precise neuromodulation of human deep brain circuits. Nature Communications, 16 (1). https://doi.org/10.1038/s41467-025-63020-1
- OBRIENJR, W. (2007). Ultrasound–biophysics mechanisms ☆. Progress in Biophysics and Molecular Biology, 93 (1-3), 212–255. https://doi.org/10.1016/j.pbiomolbio.2006.07.010
- O’Reilly, M. A., & Hynynen, K. (2012). Blood-Brain Barrier: Real-time Feedback-controlled Focused Ultrasound Disruption by Using an Acoustic Emissions–based Controller. Radiology, 263 (1), 96–106. https://doi.org/10.1148/radiol.11111417
- Rabut, C., Yoo, S., Hurt, R. C., Jin, Z., Li, H., Guo, H., Ling, B., & Shapiro, M. G. (2020). Ultrasound Technologies for Imaging and Modulating Neural Activity. Neuron, 108 (1), 93–110. https://doi.org/10.1016/j.neuron.2020.09.003
- White, P. J., Clement, G. T., & Hynynen, K. (2006). Longitudinal and shear mode ultrasound propagation in human skull bone. Ultrasound in Medicine & Biology, 32 (7), 1085–1096. https://doi.org/10.1016/j.ultrasmedbio.2006.03.015
Other Course Work
📊 Assignment 1: EEG Analysis
📓 Copy of assignment1_eeg_filtering - Samir Jaber.ipynb
Jupyter Notebook💡 Opens in Google Colab for interactive execution. Requires Google account.
🎨 Assignment 2: BrainImation
📄 Psych 403 Assignment 2 Write up - Samir Jaber.pdf
View Original PDFAssignment 2: BrainImitation Reflection
This project focused on creating a neurofeedback visualization that reflects real-time changes in
brain activity associated with stress and relaxation. Using the BrainImation platform, which
streams live brainwave data, I developed a clean and dynamic visual that responds to shifts in
mental state. The visualization displays four main brainwave bands in separate lanes that scroll
horizontally across the screen. Each waveform updates continuously, allowing users to see
fluctuations in their mental activity. The design aims to provide an immediate and intuitive sense
of how calm or tense the user is at any given moment. When the user is relaxed, the
background cools in colour and the waves appear smoother, while periods of higher stress
make the colours warmer and the motion more pronounced.
The program was built in JavaScript using the p5.js library within the BrainImation environment.
This setting automatically provides a canvas and real-time data feed, making it ideal for
responsive visualizations. The system reads the incoming values and uses smoothing to avoid
sudden jumps or flickering. Each band is drawn with a unique colour using curved lines that
animate across the screen. The strength of each band is compared to the total brain activity to
calculate two key measures: relaxation and stress levels. These values influence the
background hue and brightness, creating a colour shift that reflects the user’s changing state.
The visualization also includes an interactive feature that lets the user select a target wave by
pressing the T key or clicking on a lane, highlighting that wave for focused feedback training.
The main challenges involved stabilizing the motion of the waves and making the feedback
meaningful without overwhelming the viewer. The incoming data can be noisy, which at first
caused the animation to flicker or move erratically. Applying a smoothing function helped to
reduce that problem while keeping the response real time. Another challenge was performance.
Drawing several dynamic curves each frame required careful management of the history length
and stroke settings to maintain smooth animation. It also took experimentation to find the right
way to balance the background colour changes so that they felt natural and matched the user’s
state without becoming distracting. Lastly, I encountered AI help troubles where the AI would
constantly input “CreateCanvas” lines which would cause the screen to just appear black.
Removing these lines allowed for smooth operation and visualization of the code.
Through this project I learned how to connect scientific signals to interactive visual design. I
gained experience with real-time data visualization, signal smoothing, and user interaction in
p5.js. More importantly, I learned how small visual cues, such as gradual colour shifts or subtle
motion changes, can make self-regulation feedback both intuitive and engaging. This project
deepened my understanding of how technology, creativity, and neuroscience can come together
to support mindfulness and self-awareness.
Code:
// Simple Neurofeedback Wave Visualization + Target Toggle + Stress vs Relax response
// Lanes do not overlap, light smoothing, clean labels
// Press T to cycle target band, click a lane to set target
let waveHistory = [];
let bands = ["alpha", "beta", "theta", "delta"];
let hues = [300, 180, 60, 0]; // HSB hues for the bands
let smooth = { alpha: 0, beta: 0, theta: 0, delta: 0 };
let smoothAmt = 0.85; // higher means smoother
let targetIndex = 0;
// new, live indices
let stressLevel = 0; // 0..1
let relaxLevel = 0; // 0..1
function setup() {
colorMode(HSB, 360, 100, 100);
background(0, 0, 10);
}
function draw() {
// 1) Read current EEG. Assume values in 0..1. Fallback to simple noise if missing.
let reading = getReading();
// 2) Light smoothing
for (let b of bands) {
smooth[b] = lerp(reading[b], smooth[b], smoothAmt);
}
// 3) Compute stress and relaxation from smoothed bands
computeAffectLevels(smooth);
// 4) Background tint, warm with stress, cool with relaxation
// mix warm red,0 with cool blue,240 based on relaxLevel, then raise brightness with stress
let baseHue = lerp(0, 240, relaxLevel); // 0 is warm, 240 is cool
let baseSat = 20 + 30 * (1 - relaxLevel); // a bit more saturated when less relaxed
let baseBri = 10 + 25 * stressLevel; // brighter when stressed
background(baseHue, baseSat, baseBri);
// 5) Update history, cap size
waveHistory.push({ ...smooth });
if (waveHistory.length > width / 2) waveHistory.shift();
// 6) Draw lanes and waves, with emphasis driven by stress or relax
drawLanesAndWaves();
// 7) HUD
noStroke();
fill(0, 0, 90);
textAlign(LEFT, TOP);
textSize(16);
text("Wave Visualization, real time", 12, 10);
textSize(12);
text("Target, " + bands[targetIndex].toUpperCase() + " press T or click a lane", 12, 30);
// 8) Tiny affect readout
drawAffectHUD();
}
// ---------- input ----------
function keyPressed() {
if (key === "t" || key === "T") {
targetIndex = (targetIndex + 1) % bands.length;
}
}
function mousePressed() {
// set target by lane click
let laneH = height / bands.length;
let i = floor(mouseY / laneH);
if (i >= 0 && i < bands.length) targetIndex = i;
}
// ---------- affect logic ----------
function computeAffectLevels(b) {
// Normalize by total power, avoid divide by zero
const total = max(0.0001, b.alpha + b.beta + b.theta + b.delta);
// Relax favors alpha relative to the rest
const relaxRaw = b.alpha / total;
// Stress favors beta relative to the rest
const stressRaw = b.beta / total;
// Gentle mapping into 0..1, center near typical values
// tweak the constants if your headset yields different distributions
relaxLevel = constrain(map(relaxRaw, 0.15, 0.45, 0, 1), 0, 1);
stressLevel = constrain(map(stressRaw, 0.10, 0.40, 0, 1), 0, 1);
}
function drawAffectHUD() {
// small bars at top right
const x0 = width - 180;
const y0 = 12;
const w = 150;
const h = 8;
// labels
fill(0, 0, 85);
textAlign(LEFT, TOP);
text("Relax", x0, y0);
text("Stress", x0, y0 + 18);
// relax bar, cool hue
noStroke();
fill(220, 60, 90);
rect(x0 + 50, y0 + 2, w * relaxLevel, h);
// stress bar, warm hue
fill(10, 80, 90);
rect(x0 + 50, y0 + 20, w * stressLevel, h);
}
// ---------- helpers ----------
function getReading() {
if (typeof eegData !== "undefined") {
return {
alpha: constrain(eegData.alpha ?? 0, 0, 1),
beta: constrain(eegData.beta ?? 0, 0, 1),
theta: constrain(eegData.theta ?? 0, 0, 1),
delta: constrain(eegData.delta ?? 0, 0, 1)
};
}
// simple simulation if no EEG present
let t = millis() * 0.0008;
return {
alpha: 0.5 + 0.3 * noise(t * 1.2),
beta: 0.4 + 0.3 * noise(t * 1.6 + 10),
theta: 0.5 + 0.2 * noise(t * 0.9 + 20),
delta: 0.35 + 0.2 * noise(t * 0.6 + 30)
};
}
function drawLanesAndWaves() {
let laneH = height / bands.length;
let amp = laneH * 0.4; // vertical amplitude inside each lane
textSize(12);
textAlign(LEFT, CENTER);
for (let i = 0; i < bands.length; i++) {
let band = bands[i];
let hue = hues[i];
// lane center
let cy = laneH * i + laneH * 0.65;
// background highlight for target lane
if (i === targetIndex) {
noStroke();
fill(0, 0, 20, 30);
rect(0, laneH * i, width, laneH);
}
// lane baseline
stroke(0, 0, 25);
strokeWeight(1);
line(0, cy, width, cy);
// label
noStroke();
fill(0, 0, 80);
text(band.toUpperCase(), 10, cy - laneH * 0.45);
// adaptive emphasis
// alpha gets stronger with relaxation, beta gets stronger with stress
let weightBoost = 0;
let briBoost = 0;
if (band === "alpha") {
weightBoost = 2 * relaxLevel; // up to +2 px
briBoost = 20 * relaxLevel; // brighten with relax
} else if (band === "beta") {
weightBoost = 2 * stressLevel;
briBoost = 20 * stressLevel;
}
// target lane is always a tad thicker
const baseWeight = i === targetIndex ? 3 : 2;
const strokeW = baseWeight + weightBoost;
// wave
stroke(hue, 80, min(95 + briBoost, 100));
strokeWeight(strokeW);
noFill();
beginShape();
for (let k = 0; k < waveHistory.length; k++) {
let x = map(k, 0, waveHistory.length - 1, 0, width);
let v = waveHistory[k][band]; // already 0..1
let y = cy - v * amp; // draw upward within lane
vertex(x, y);
}
endShape();
}
}
🧠 Interactive Neurofeedback Visualization
💡 Code is embedded in this portfolio - opens instantly in the live BrainImation editor. Press T to cycle target bands, click lanes to select!
🎥 Screen Recording 2025-10-17 at 11.43.56 AM - Samir Jaber.mov
💡 Videos require Google Drive access. Open in new tab if it doesn't load.
🎯 Midterm Project
📄 Jaber_PSYCH 403_Midterm Report - Samir Jaber.pdf
View Original PDFSamir Jaber
1755507
Midterm Report
Part 1:
For this project, I implemented a Target Brain State Trainer that provides real-time
neurofeedback rather than simply mapping EEG values to colors or motion. The system
monitors alpha and theta activity and continually compares the ratio between them to a
user-defined target associated with relaxed focus. It measures the difference from that target
each frame, converts the distance into a 0-to-1 progress score, and gradually rewards stable
control through an accumulating point total. Visual feedback comes through a dynamic
background hue and saturation that brighten and shift as the user’s brain activity approaches
the goal. The interface also detects eyes-closed periods using elevated alpha power and
displays this classification in the HUD.
The control loop is continuous:
Brain waves to Alpha/Theta ratio to Error from target to Visual and score feedback to User
adjusts mental state to Brain waves update. Because feedback and error correction occur every
frame, the participant learns to self-regulate brain rhythms in real time, producing a closed
adaptive loop rather than a one-way mapping.
This approach differs from basic parameter control because the visualization no longer acts as a
passive display. It quantifies performance, adapts to the user, and drives behavioral learning. A
calibration key lets the user set a personalized target ratio, while adjustable brackets fine-tune
difficulty. The program’s visual harmony emerges only when the brain enters the intended state,
effectively turning the artwork into a training instrument for relaxation and attention balance.
Implementing this taught me how neurofeedback systems translate continuous EEG signals into
meaningful behavioral cues. I learned the importance of smoothing, normalization, and
incremental scoring to prevent noise from misleading the feedback loop. Most importantly, I
experienced how real-time feedback can make the invisible rhythms of the brain perceptible and
trainable through art and interaction.
Part 2:
For this ERP experiment, I implemented a P300 Oddball Paradigm to measure the P300
component, a large positive deflection that typically occurs around 300 ms after a person
detects an infrequent or meaningful event. The P300 is strongly linked to attention and cognitive
evaluation processes, making it a central target in event-related potential research.
In this experiment, colored circles represent two stimulus types: frequent standard stimuli (blue)
and infrequent oddball stimuli (red). Approximately 20% of the trials are oddballs, and each rare
event is accompanied by a brief 880 Hz beep to enhance saliency and simulate an
attention-driven task. Participants view one stimulus per second, with each presentation
followed by a short inter-trial interval that allows the signal to return to baseline.
The system also includes several keyboard controls for flexibility during testing: S starts or
pauses the experiment. R resets all averages and restarts the trial count. M toggles simulation
mode, allowing testing without a live EEG feed. [ and ] adjust the rare stimulus probability (from
5% to 90%). + and – change the vertical scale of the ERP plots. A manually plays the
rare-stimulus beep for testing audio playback.
The averaging system continuously monitors the EEG data stream (or simulated signal) and
marks each stimulus onset. It then extracts epochs from −200 ms to +800 ms relative to the
event, performs baseline correction using the pre-stimulus interval, and averages up to 20
epochs per condition. The averaged traces for standard and rare trials update in real time,
allowing users to watch the ERP waveforms emerge as noise cancels out through repeated
averaging.
If tested with real EEG data, the rare condition would produce a clear P300 component, a strong
positive peak around 300 ms post-stimulus, especially over parietal electrodes. Standard trials
would primarily show early sensory activity without this late positivity, replicating the classic
pattern described in decades of P300 research.
🎥 403 Midterm Part 2 - Samir Jaber.mov
💡 Videos require Google Drive access. Open in new tab if it doesn't load.
🎥 403 Midterm Part 1 - Samir Jaber.mov
💡 Videos require Google Drive access. Open in new tab if it doesn't load.
📝 Jaber_midterm_part2 - Samir Jaber.txt
💡 Code is embedded in this portfolio - opens instantly in the live BrainImation editor (no internet required!)
📝 Jaber_midterm_part1 - Samir Jaber.txt
💡 Code is embedded in this portfolio - opens instantly in the live BrainImation editor (no internet required!)