The physical trick behind fUS is ultrafast plane-wave imaging. Instead of scanning tissue line by line as conventional ultrasound does, fUS transmits plane or diverging waves and reaches frame rates approaching several thousand per second, near the theoretical physical limit. Sophisticated clutter filtering then separates the slow motion of blood cells from the faster motion of tissue, exposing vascular hemodynamic changes that ordinary ultrasound cannot see. The result is a spatial resolution of roughly 100 micrometers, sensitivity sufficient to detect flow changes in vessels smaller than 100 micrometers in diameter, and a package that is portable, free of ionizing radiation, and comparatively cheap.

The biological foundation is neurovascular coupling, the precisely regulated mechanism by which active neurons recruit their own blood supply. When neural ensembles fire, they release glutamate, which activates astrocytic endfeet, propagates calcium waves, and triggers the release of vasoactive substances. Local cerebral blood flow dilates and rises two- to four-fold within seconds. Following sensory stimulation, local cerebral blood volume begins rising within 0.5 to 2 seconds, peaks at about 4 to 6 seconds with an amplitude increase of 10 to 50 percent over baseline, and returns to rest within 8 to 15 seconds. Because the fUS signal scales approximately linearly with local firing rate, tracking this blood-volume profile in real time provides an indirect but faithful readout of population-level neural activity, robust enough for single-trial decoding without the averaging that fMRI typically demands.

The field’s pivotal moment came in 2021, when a California Institute of Technology-led team published in Neuron the first demonstration that fUS could decode motor intentions from the posterior parietal cortex of non-human primates through an epidural probe and an ultrasound-transparent skull window, avoiding direct contact with brain tissue. Offline decoding of movement intentions across eight directions exceeded 70 percent accuracy, and fUS captured signals from the primary motor cortex that distinguished flexion intentions of different fingers. In 2024, the same collaboration reported in Nature Neuroscience the first complete closed-loop demonstration on the read side, with epidural fUS decoding of movement plans driving a brain-machine interface in real time. Online decoding from preparatory-period signals achieved 75 to 82 percent accuracy for two directions and 45 percent for eight, with end-to-end latency near 800 milliseconds.

Clinically, the review sorts applications by maturity. The most human-ready use is neonatal epilepsy monitoring: fUS has already imaged seizure-related hemodynamic changes through the unclosed anterior fontanelle of human newborns, a Tier 1 application with demonstrated feasibility. Stroke assessment sits at Tier 2, supported by small exploratory studies of fewer than 50 patients. Because acute ischemia severely impairs neurovascular coupling in the penumbra, where residual neural activity fails to trigger blood flow increases, fUS can monitor tissue viability in real time, and one NVC-guided endovascular treatment group showed significantly better functional outcomes than conventional imaging guidance. Early screening for Alzheimer’s and Parkinson’s disease rounds out the portfolio, with hippocampal NVC function in AD patients positively correlating with cognitive scores, and super-resolution ultrasound localization microscopy detecting capillary density changes in AD mouse models before structural deficits appear.

The frontier is bidirectionality. Transcranial focused ultrasound neuromodulation uses low-frequency ultrasound below 1 MHz to reach deep nuclei such as the thalamus and hippocampus with millimeter-scale targeting, and it has been applied to essential tremor, Parkinson’s disease, treatment-resistant depression, and obsessive-compulsive disorder. Sonogenetics, which emerged as a breakthrough in 2024, expresses ultrasound-sensitive proteins such as Piezo1 and MscL in specific neurons so that ultrasound can activate them with molecular precision, and nanobubble-actuated neuromodulation has achieved sub-millimeter shaping of behavior in mice. Four-dimensional fUS, demonstrated in 2019 by French institutions, captures whole-brain network activity in three-dimensional space plus time. Together these technologies are pushing fUS from single-signal acquisition toward acquisition-modulation closed-loop systems, though the review is explicit that no integrated read-write demonstration yet exists in animals or humans.

Safety looms as the field’s most sobering question. A 2025 preprint in squirrel monkeys reported that repeated high-frequency transcranial focused ultrasound at 1.8 MHz induced neurodegeneration, detected as Fluoro-Jade C-positive degenerating neurons, even though the estimated peak temperature rise was only 0.13 degrees Celsius and the mechanical index stayed below the FDA limit. Degeneration scaled with cumulative active exposure time rather than peak intensity. Importantly, the review notes that this exposure paradigm, with 150-millisecond pulse trains repeated every 10 seconds for up to 90 minutes, is orders of magnitude more intense than the diagnostic imaging regime of fUS-BCI, whose individual frames last roughly 0.1 to 0.5 milliseconds at a mechanical index around 0.1. Still, the finding underscores that repeated long-duration paradigms are not covered by current guidelines, and the review recommends the ALARA principle with mechanical index below 0.5 and thermal index below 1.0.

The skull remains the central engineering obstacle, blocking roughly 90 percent of ultrasound signal and creating what the review calls an impossible triangle among non-invasiveness, high spatial resolution, and real-time capability. Microbubble contrast agents can boost vascular signals nine-fold but circulate in the brain for only 2 to 5 minutes, while ultrasound localization microscopy achieves capillary-level resolution but requires minutes of processing that fails real-time BCI requirements. Adaptive beamforming algorithms can correct skull-induced aberrations, but patient-specific modeling is computationally demanding. In the near term, implantable ultrasound-transparent windows made of materials such as TPX may represent the optimal transitional solution. Hardware is shrinking fast: Forest Neurotech and Butterfly Network developed an ultrasound-chip BCI, and the Forest 1 research-grade scanner launched in 2025 is smaller than a standard key fob, while researchers in China created an injectable 2-millimeter supergel sensor that wirelessly monitors intracranial signals without battery or RF architecture.

Three bottlenecks demand urgent resolution. The hemodynamic response latency of 1 to 2 seconds limits scenarios requiring millisecond timing, though predictive decoding of preparatory motor signals can achieve effective response times in the hundreds of milliseconds, and hybrid architectures fusing fUS with EEG could supply millisecond-scale temporal predictions alongside spatially precise goal-state updates. Under pathological conditions, neurovascular coupling can be attenuated, delayed, or dissociated from neural activity, so decoders calibrated on healthy brains misestimate intention in proportion to local uncoupling; per-patient baseline calibration, concurrent electrophysiology, and physiologically constrained algorithms offer mitigation. There are also regulatory and ethical gaps: fUS equipment costs roughly 300,000 to 500,000 euros per unit, existing frameworks were built for electrophysiological BCIs, and fUS’s spatial resolution enables decoding of specific mental content that raises mental privacy and neural data sovereignty concerns beyond EEG’s reach.

The review closes with a concrete roadmap. Over the next 3 to 5 years, priorities include constructing large-scale pathological NVC databases across stroke, epilepsy, and Parkinson’s disease; establishing chronic implantable ultrasound safety standards with multidimensional thermal, mechanical, and cumulative dosage thresholds; conducting multicenter randomized controlled trials validating efficacy in motor rehabilitation and seizure prediction; and advancing CMUT scalable manufacturing to cut costs by one to two orders of magnitude. Looking ahead 5 to 10 years, the author expects fUS-BCI to achieve millisecond-scale neural intent decoding and submillimeter closed-loop modulation of deep brain nuclei, transforming neurological care from experience-driven practice toward data-driven closed-loop medicine. For now, the technology remains at the proof-of-concept stage, but the convergence of ultrafast imaging, adaptive algorithms, and chip-scale integration suggests that ultrasound may soon become the precision interface the BCI field has been waiting for.

Subject of Research: Functional ultrasound brain-computer interfaces based on neurovascular coupling for neurological diseases

Article Title: Applications of functional ultrasound brain-computer interfaces in neurological diseases: research progress based on neurovascular coupling

Article References: Applications of functional ultrasound brain-computer interfaces in neurological diseases: research progress based on neurovascular coupling. (n.d.). https://doi.org/10.3389/fnins.2026.1946753

Image Credits: AI Generated

DOI: 10.3389/fnins.2026.1946753

Keywords: functional ultrasound, brain-computer interface, neurovascular coupling, neurological diseases, stroke, epilepsy, Alzheimer's disease, neuromodulation, sonogenetics, ultrafast Doppler imaging, neurorehabilitation, biosafety