fUSI in 10 questions¶

This section is an introduction to fUSI in the form of 10 questions to get you started for a successful fUSI project. These questions will serve as an introduction to the main topics discussed in this website.

0. What is fUSI?¶

Let us start with the most basic of all questions: What are we talking about? In a nutshell, fUSI is a relatively new imaging modality which allows to image the brain activity through its vascular dynamics, the so-called hemodynamic signal. Using advances in the field of acoustics, physicists have been able to increase significantly the frame rate of ultrasound imaging. By doing so they unlocked the possibility to track evolution of cerebral blood volume in time with an ultrasound probe at a resolution of ~100 μm x 100 μm x 400 μm x 0.5 s. fUSI usually relates to Power Doppler ultrafast ultrasound imaging which gives an indirect access to neuronal activity thanks to the neurovascular coupling. We can now also observe the vasculature at the scale of ~5 μm x 5μm x 5μm with Ultrafast Localisation Microscopy (ULM), a non-linear imaging technique that was recently extended in the time domain with fULM. This method can also be used to measure mechanical properties of the brain like strain or elastography. As the later examples are still early in their development, here we mainly focus on Power Doppler imaging, with some sections describing ULM more specifically.

1. What is your research question?¶

fUSI can be applied to so many fields (cardiology, nephrology, hepatology…), but in this website you will mainly find information related to neuroimaging. Since fUSI relies on measuring blood volume, there are several fields of research where it can prove useful. On the pure vascular aspect, Ultrafast Localisation Microscopy (ULM) is probably the most relevant tool. If you are interested in neurovascular coupling, fUSI is a great choice and can easily be paired with controlled sensory stimulations, or even direct electrophysiological recordings. Finally fUSI can provide an indirect measure of neuronal activity, and can be used to study large-scale patterns of brain activity in resting-state paradigms, in relation to sensory stimuli, behavioral tasks or events, or following perturbational approaches like optogenetics or chemogenetics.

2. Which hardware do you need?¶

Any fUSI experiment will require dedicated machines. The minimal rig should always need:

A ultrasound system with a probe (the machine that sends and receives ultrasound pulses)
A computer to process the massive amount of data generated by ultrafast imaging

A this stage you have to decide if you go for a custom system, which requires a bit of knowledge in wave physics, ultrasound imaging and engineering, but starts to benefit from open source resources. Such a system will allow you to control every parameters of the experiment, which can be extremely time-consuming to benchmark with crazy factorial designs in the parameter. Fortunately, the field has progressed in the last decades and typical parameters are now broadly used. This allowed for some commercial solutions to emerge that can make experimental work much easier without having to think too much about hardware and data acquisition. So once more, depending on your question you might have to favor one or the other option.

More on that topic here.

3. Which probe should you use?¶

For decades ultrasound imaging has been dominated with linear arrays of transducers. Such probes allow to image a single slice and were motivated by both the focused insonification and limited computational resources. With the emergence of plane wave imaging and the exponential increase of online processing capacities, the number of element in arrays has increased to allow modular imaging with wavefront patterning in emission and in silico beamforming in reception. As a result, multiple probes are now available for fUSI acquisitions allowing different settings from 2D snapshots to 3D+T fast movies. The choice of your probe will influence the sensitivity, spatial resolution, spatial coverage and temporal resolution you can reach with your setup.

More on that topic here.

4. Which animal model do you want to use?¶

fUSI has been used in many species: mice, rats, guinea pigs, ferrets, rabbits, pigeons, crows, Australian bearded lizards, sheep, pigs, monkeys, babies, humans (non-exhaustive list!). Depending on the animal model, some invasive surgeries might be required to image the brain or the spinal cord (see question 5). The advantage of small rodents is that almost the entire brain can be imaged at once. For bigger brains, the current probes can only cover a limited field of view and depth (depending on the probe), but a given region will contain more voxels, allowing for different investigations such as fine-grained sensory representations. For monkeys or humans, it is usual to use a lower frequency like 8 MHz as opposed to 15 or even 40 MHz in mice. A tradeoff must be found as the lower the frequency, the higher the penetration depth (respectively 1 cm to >10 cm) but the lower the spatial resolution (respectively 40 μm to 200 μm).

5. What preparation for your animal model?¶

The skull that protects fragile brain tissue from environmental hazards is a blessing for the brain but a nightmare for neuroimagers. For ultrasounds, the bone is a random dissipater that attenuates and scatters the acoustic wave, resulting in imaging aberrations. Consequently, fUSI applications are mainly limited by invasiveness. So far non invasive imaging has only be reported in mice (up to a few months old), young rats (a few weeks old), and human babies through the fontanelle. In other cases, a cranial imaging window is usually required. In rats, skull thinning (removal of the first two layer of the bone) has been proved to give high quality signal. For other models, the most common method consists in replacing the skull by a material transparent for ultrasound (usually polymethylpentene: PMP or TPX). Alternatively, a contrast agent (inert microbubbles or gas vesicles) can be used to image through the skull, but this might result in signal inhomogeneity. This technology is the foundation of ULM. Finally, for acute preparations and intra-operative acquisitions, fUSI can be imaged directly on the brain after a simple craniotomy.

More on that topic here.

6. Anesthesia or awake?¶

The choice of performing anesthesia or awake recordings depends on multiple parameters to consider based on your specific experimental needs.

The main advantage of anesthetized recordings is the absence of movement from the animal, which drastically decreases movement artefacts. Indeed, movement is literally the basis of the measured signal in fUSI, so that contamination by artefactual movement is frequent. As for fMRI, movement artefacts fill the nightmares of the fUS imager. It is interesting to notice that with a new modality, the same problem has completely different symptoms. Some movements (big amplitude but slow dynamics) can be dealt with with conventional motion correction algorithms from the MRI community. Other types of movements, such as typical local drifts (faster and lower amplitude, typically sub-voxel) found in BOLD signals because of magnetization inhomogeneity, are almost absent in fUSI. Conversely, sub-voxel displacement of the entire brain associated with movement will propagate a sharp increase in power Doppler across the entire brain, as far out as the gel itself which otherwise bears no signal. The transition form anesthesia to awake is therefore non trivial as the signal signature varies significantly.

Thus, while transcranial preparations are well combined with anesthesia in mice, in awake animals, a craniotomy may be needed to increase the sensitivity and therefore signal to noise ratio by orders of magnitude, significantly reducing the influence of such artefacts. Each of these settings also have additional drawbacks to consider: anesthesia also directly affects brain activity as well as possibly neurovascular coupling, which might bias scientific results; craniotomies reduce the coverage and complexify the experimental protocol; transcranial imaging can only be done during a few months before the skull thickens and signal quality decays.

More on that topic here.

7. What protocol for your study?¶

Now that you know your model and have the right machine with the right preparation, you can think of an experimental plan. There are infinite possibilities, but let’s list here the most usual experimental paradigms with fUSI. Note that most of these paradigms will need a system to synchronize fUSI data with other variables (for e.g. stimuli timings, behavioral measurements, task events).

Stimulation paradigms: subjects can be presented with any kind of sensory input: visual with various flavors (LED, screens, movies) and various orientation for retinotopy studies or oculomotor reflexes, auditory from simple tones to natural sounds, olfactory down to the stimulation of single glomeruli or tactile, from whisker stimulations to electric shocks. The effects of these stimuli can then be analyzed with PSTHs, GLMs or any analysis method that you fancy.
Task paradigms: fUSI is in principle compatible with the study of animals engaging in behavioral tasks and has successfully been used for this purpose, for example in ferrets performing an auditory-based decision-making task, monkeys performing memory-guided saccades, or rats engaging in freely-moving locomotion. These paradigms are more challenging from a processing point of view as motion artefacts are likely to be stronger, and multiple events may overlap in time at scales that cannot be separated by fUSI.
Resting state paradigms: it surprisingly consists in leaving the subject doing nothing while recording the brain activity. These study are targeting intrinsic activity of the brain within frameworks like functional connectivity arousal fluctuations. Given the massive impact of behavioral state on brain activity, it is recommended to simultaneously record as many behavioral and physiological parameters as possible, such as facial movements, whisking, pupil and eye position and size, locomotion, breathing, skin conductance or heart rate.
Perfusion fluctuation paradigms: while the previous methods focused on changes in blood volume at the scale of a few seconds, slower fluctuations in blood volume can also be investigated with fUSI. For tracking systemic flow or global perfusion after a stroke, the brain can be imaged much slower at a high resolution/sensitivity (using the typical tradeoff of imaging that the longer the acquisition better the signal), for example with an image every minute or so.
Ultrafast Localization Microscopy (ULM): for ULM, micro bubbles are injected in the blood stream through a vein, and then tracked within the brain. Under the constant hematocrit hypothesis, accumulation of bubble distribution in the brain allows to estimate quantities like vessel diameters, average blood flow vector and derivatives at the 5 μm resolution. One such image require few minutes to accumulate enough information on the bubbles distribution.

8. What processing for your data?¶

Now that you acquired your data you are entering the fantastic realm of data analysis. Fortunately, you can now rely on generations of neuroscientists who identified very smart ways to make sense of brain signals independently of all the noise it usually comes with. There are no limits here, only endless possibilities. There is no absolute consensus on how to best process your data, and this remains an active line of research. However, some common practices have emerged, which we’ll briefly cover here:

Standardization of your data (which is the topic of our last question)
Registration of the data which consists in warping individual images on a reference template or the other way around if you analyze ROI timeseries.
Processing/denoising of timeseries will be preparation-, protocol-, state- and dataset-dependent but might include some of the following steps:
- Motion correction: intra-recording registration which aligns all the frames together to correct for potential movement of the brain during the acquisition
- Filtering: whether it is spatially or temporally most pipelines apply some filter to separate noise from signal of interest
- Regression: which removes expected sources of noise (extracted with your favorite method: PCA, ICA, CCA… You name it)
- Scrubbing/censoring: the last resort: removing frames you know contain artefacts from your timeseries, and possibly interpolate to fill the missing values, depending on your following analysis.
Dimension reduction, to reduce your high-dimensional recording into a simpler observable related to your question: like an activation map for a stim, a correlation map for FC, stimulation average timeseries, etc. This is the topic of the next and last question.
Population level statistics is an optional but recommended step to confirm that the effect that you report is representative of a population.

More on that topic here.

9. What analysis for your data?¶

… And the need to pilot them.

This is the final step of your fUSIing experience… But funnily enough, it shall be the first too. Here we come back to the scientific question you want to address. The analysis you perform is highly constrained by the way you acquired your data. As a result it is essential that you get a precise idea of the type of analysis you want to apply for it before starting the experiments. If Exploratory Data Analysis (EDA) is a good way to discover your data, it should be limited to your pilot studies avoiding any circularity and overfitting of your tools to your dataset. It is why a proper pilot study should encompass analysis steps as well. It usually is the moment you realize about synchronization problems, too deep anaesthesia levels, and other discrepancies. Another important point which impacts fUSI is the neuronavigation and registration issues. You should make sure before acquiring a dataset that you know where you are in the brain, and that you have an efficient way to register your data with your specific sensitivity/resolution. Failing to do so increases significantly the risk of acquiring datasets unfit to address your question (and there might be some painful experiences motivating these lines). We can not list here all the analyses you can do (among which you’ll find your favorite acronym: GLM, PCA, CPCA, ICA, CCA, CHARM, fALFF, SBA, AM, tSNE…), but you can find below a link where we will try to list as many as we can, and help you implement them.

More on that topic here.

With these ideas in mind, you should be better equipped to begin your experiments — whether you’re a newcomer gaining a clearer understanding of the field you’re stepping into, an experienced user discovering new perspectives for your next experiments, or a seasoned veteran ready to share valuable feedback with the community. We sincerely hope you can find comfort and support in the resources available here, as you will face the many challenges of this exciting and ever-evolving field. And if you have more questions, you know where to ask 😉