🔬 physics

Automatic Aberration Correction for Transcranial Functional and Super-Resolution Ultrasound Imaging in Rodents and Nonhuman Primates

This paper presents a fully automated, differentiable beamforming framework that effectively corrects skull-induced aberrations to significantly enhance the resolution, sensitivity, and accuracy of transcranial functional and super-resolution ultrasound imaging in both rodents and nonhuman primates.

Original authors: Paul Xing, Antoine Malescot, Eric Martineau, Stephan Quessy, Ravi L. Rungta, Numa Dancause, Jean Provost

Published 2026-05-05

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Paul Xing, Antoine Malescot, Eric Martineau, Stephan Quessy, Ravi L. Rungta, Numa Dancause, Jean Provost

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine trying to take a crystal-clear photo of a bustling city street, but you have to do it through a thick, wavy, and uneven piece of glass. The glass distorts the image, making the buildings look stretched, the cars look like they are in two places at once, and the fine details of the streetlights blur into a mess. This is exactly what happens when scientists try to use ultrasound to look inside a living brain through the skull. The skull acts like that wavy glass, scrambling the sound waves and ruining the picture.

This paper introduces a clever new "digital lens" that automatically fixes these distortions, allowing for incredibly sharp, high-resolution images of the brain's tiny blood vessels without needing to cut open the skull.

Here is how the researchers solved the problem, broken down into simple concepts:

The Problem: The "Skull Blur"

Ultrasound Localization Microscopy (ULM) is a super-powerful technique that tracks tiny bubbles (microbubbles) flowing through blood vessels. It's so good it can see vessels as thin as a human hair. However, when you try to do this through a skull (in mice, monkeys, or humans), the bone bends the sound waves.

The Result: The image gets blurry. A single blood vessel might look like two separate ones (a "ghost" image), or parts of the network might look disconnected. It's like trying to read a map through a funhouse mirror.

The Solution: A "Self-Correcting" Camera

The team created a new computer method called differentiable beamforming. Think of this as a camera that doesn't just take a picture; it constantly asks, "Is this image sharp?" and if not, it automatically adjusts its own settings until the picture is perfect.

Here is the step-by-step process they used:

The "Isoplanatic Patches" (The Grid): Instead of trying to fix the whole image at once, they divided the brain image into a grid of small squares (patches). Imagine a checkerboard laid over the brain.
The "Tuning Knobs": For each square on the checkerboard, the computer has a virtual "knob" that controls how much to delay the sound waves. Because the skull is uneven, different parts of the brain need different amounts of "tuning."
The "Coherence Test" (The Judge): The computer uses a metric called angular coherence. Think of this like a choir. If everyone sings the exact same note at the exact same time, the sound is powerful and clear (high coherence). If they are slightly out of sync, the sound is muddy. The skull makes the sound waves "out of sync." The computer's goal is to turn the knobs until all the sound waves sing in perfect harmony again.
The "Auto-Tune" Loop: The computer uses a mathematical trick (gradient ascent) to automatically tweak those knobs. It checks the "choir," sees they are out of tune, adjusts the knobs, checks again, and repeats this thousands of times in seconds until the image is perfectly focused.

What They Found

The researchers tested this on mice and non-human primates (macaques).

Sharper Images: Before the fix, blood vessels looked fuzzy or duplicated. After the fix, the vessels were crisp, single lines. In fact, the resolution improved significantly, allowing them to see vessels that were previously invisible or broken.
Fixing "Ghost" Vessels: They successfully removed the "ghost" images where one vessel looked like two. The corrected images showed the true, continuous path of the blood flow.
Better Blood Flow Data: Because the image was clearer, they could measure blood speed more accurately. Before the fix, the blood flow looked chaotic and turbulent (like a stormy river). After the fix, it looked smooth and laminar (like a calm river), which is how blood actually flows in a healthy brain.
Functional Imaging: They also used this to study brain activity. When they stimulated a mouse's whiskers, the corrected images showed a much clearer and stronger signal in the brain areas responsible for feeling that touch, including deep areas that were previously hidden by the skull's distortion.
3D Success: They even managed to do this in 3D for the monkeys, correcting distortions in depth as well as side-to-side, proving the method works for complex, real-world brain shapes.

Why This Matters (According to the Paper)

The paper claims this is a major breakthrough because it is fully automatic. You don't need a special guide star (like a specific type of bubble) or a pre-scan of the skull to make it work. It works on standard data from mice and monkeys.

By removing the "skull blur," this technology lowers a huge barrier. It means scientists can finally get high-definition, super-resolution maps of brain blood flow and function without needing to drill holes in the skull, making it possible to study the brain in a much more natural, non-invasive way across different species.

Technical Summary: Automatic Aberration Correction for Transcranial Functional and Super-Resolution Ultrasound Imaging

Problem Statement
Transcranial ultrasound localization microscopy (ULM) and functional ultrasound (fUS) offer the potential for non-invasive, super-resolution visualization of brain microvasculature and hemodynamics. However, the skull introduces significant acoustic aberrations that degrade image quality. These aberrations reduce microbubble detection sensitivity, lower contrast and resolution, and cause critical artifacts such as vessel duplication, disconnected vessels, and biased velocity estimations. While previous correction methods exist, they often rely on specific guide stars (isolated microbubbles), require extensive datasets (e.g., full synthetic aperture), are limited to larger vessels, or necessitate invasive cranial windows. Furthermore, deep learning approaches often struggle to generalize across different experimental settings without extensive retraining.

Methodology
The authors propose a differentiable beamforming framework for automatic, closed-loop aberration correction that does not require prior training or a separate calibration dataset. The core methodology involves:

Differentiable Beamforming: Implemented using PyTorch, the system treats the ultrasound beamforming process as a differentiable computational graph. This allows for the automatic computation of gradients with respect to the aberration parameters.
Spatially Distributed Parameterization: Instead of assuming a uniform phase screen, the skull-induced aberration is modeled as a spatially distributed function. This function is parameterized by multiple isoplanatic patches (initially set to $8\lambda \times 8\lambda$ ) covering the field of view. These patch values are upsampled via cubic interpolation to provide a continuous, pixel-wise phase correction profile.
Objective Function (Angular Coherence): The optimization goal is to maximize the angular coherence of the ultrasound signal. The system computes the angular covariance matrix ( $C_\theta$ ) from the beamformed images of different transmit angles. The first singular value ( $\sigma_1$ ) of this matrix serves as the objective function, representing the level of coherence between plane waves.
Gradient Ascent Optimization: Using automatic differentiation, the system iteratively updates the phase delays of the isoplanatic patches via gradient ascent to maximize $\sigma_1$ . The process includes L1 and Laplacian regularization to ensure smooth solutions and prevent overfitting.
Applicability: The framework operates on raw channel data after singular value decomposition (SVD) clutter filtering. It functions effectively in the presence or absence of microbubbles and is applicable to both 2D and 3D imaging configurations.

Key Contributions

Generalizable Correction: A fully automated solution that corrects aberrations without needing guide stars, specific training data, or extensive synthetic aperture acquisitions.
Closed-Loop Optimization: A physics-informed approach that optimizes focusing by directly maximizing signal coherence, adaptable to various emission schemes (plane waves, divergent waves).
Extension to 3D: Successful adaptation of the method to 3D transcranial ULM using a matrix array, capturing both lateral and depth variations of the aberration with a limited set of transmit angles.

Results
The framework was validated in vivo on rodents (mice) and nonhuman primates (NHPs, macaques):

Mouse Imaging (2D):
- Doppler & CEUS: Aberration correction improved the contrast-to-noise ratio (CNR) and signal-to-noise ratio (SNR) in both young (2-3 months) and older (6-7 months) mice, with and without microbubbles. Vessel widths were significantly reduced (e.g., from $202.4 \pm 50.2$ µm to $162.6 \pm 27.8$ µm in young mice), indicating improved resolution.
- ULM: Resolution improved from $23.8 \pm 5.4$ µm to $18.4 \pm 3.0$ µm in young mice. The number of detected microbubble tracks increased significantly, and previously disjoint or duplicated vessels were correctly connected.
- Hemodynamics: Correction recovered physiological parabolic flow profiles and reduced spurious velocity estimates. Mean track velocities decreased (indicating better reconstruction of smaller vessels), and track lengths increased, allowing for more reliable hemodynamic quantification.
- Functional US (fUS): In whisker stimulation experiments, correction revealed activation in the ventral posteromedial nucleus (VPM) of the thalamus, which was previously obscured, and increased the amplitude of the hemodynamic response.
Nonhuman Primate Imaging (2D & 3D):
- 2D: Using a 2.5 MHz phased array, the method removed vessel duplications around the Circle of Willis and improved the connectivity of lenticulostriate arteries. Resolution improved from 177 µm to 162 µm.
- 3D: Using a 3 MHz multiplexed matrix array, the method corrected major artifacts (vessel duplications) in 3D ULM. Resolution improved from 110.2 µm to 93.6 µm. The system achieved this using only 5 transmit angles, demonstrating efficiency.

Significance and Claims
The authors state that this work lowers a major technical barrier to transcranial ultrasound imaging by providing a generalizable, fully automated solution for aberration correction. Key claims include:

The method enables robust super-resolution and functional neuroimaging across species (rodents to NHPs) and imaging depths.
By correcting aberrations, the approach allows for reliable hemodynamic quantification and the recovery of physiological flow dynamics that are otherwise distorted by the skull.
The technique is particularly promising for clinical translation, as the macaque skull thickness is comparable to the human temporal bone, suggesting the method could facilitate non-invasive fUS and ULM in humans without the need for cranial windows or guide stars.
The framework represents a step toward high-resolution, non-invasive imaging of brain structure and function, potentially aiding in the diagnosis of cerebrovascular diseases and the study of neurovascular coupling.

The Problem: The "Skull Blur"

The Solution: A "Self-Correcting" Camera

What They Found

Why This Matters (According to the Paper)

More like this