Future Intelligence Bureau | Clinical Advancement of Ultrasound-Based Brain-Computer Interfaces: Reading and Writing the Brain Without Implanted Electrodes?

Source: Xinhua Finance

Shanghai, July 7 (Xinhua Finance) — Brain-computer interfaces are evolving from competition among single technical routes toward more diversified industrial exploration. In addition to the well-known implanted electrode approach, non-invasive or minimally invasive pathways based on physical methods such as ultrasound, light, and magnetism are also beginning to emerge.

Recently, Gestala announced the completion of its Angel+ round of financing, raising RMB 420 million. Instead of implanting electrodes into the brain, the company has opted for an ultrasonic brain-computer interface (BCI) technological approach.

“Neuroscience and artificial intelligence are two sides of the same coin,” said Peng Lei, founder and CEO of Gestala. The ultrasound technology pathway opens up greater possibilities for collaboration between brain-computer interfaces (BCIs) and AI: multimodal neural data from ultrasonic BCIs require AI-driven decoding and analysis to explore the brain’s underlying mechanisms. While AI helps humans better understand complex brain activities, it also provides new inspiration for the evolution of next-generation AI algorithms.

Not a Simple Distinction Between "Invasive" and "Non-Invasive"

In recent years, brain-computer interface (BCI) technology has largely focused on implantable solutions. By implanting electrodes into the brain, these devices hold promise for helping patients with conditions such as amyotrophic lateral sclerosis (ALS) and high-level paraplegia to control mice, robotic arms, or other external devices. The advantage of this approach lies in its higher signal acquisition accuracy, making it particularly suitable for motor control tasks.

However, Peng Lei believes that brain-computer interfaces (BCIs) should not be classified solely as “invasive” or “non-invasive.” He proposes understanding BCIs as a matrix: the horizontal axis represents different physical modalities such as acoustic, optical, electrical, and magnetic methods, while the vertical axis encompasses invasive, semi-invasive, and non-invasive approaches. Each cell in this technological matrix may correspond to distinct indications and product forms.

“Every niche has the potential to spawn a company, but ultimately it depends on whether there are patients who will use this technological combination,” said Peng Lei. Invasive electrode solutions have advantages in scenarios involving motor output control, such as cursor movement, and Gestala is not currently attempting to compete with them in the same domain. “We have chosen different indications. Their focus is on patients with amyotrophic lateral sclerosis (ALS) and high-level paraplegia, whereas we target pain management, psychiatric disorders, and stroke rehabilitation. The patient populations differ, and so do the ways in which they benefit.”

Gestala’s first priority indication is currently chronic pain. Peng Lei stated that the company does not believe ultrasound brain-computer interfaces will replace medications; however, for patients suffering from long-term pain, those at high risk of drug dependence, or those unsuitable for long-term, high-dose medication due to complications, non-surgical neuromodulation may offer new therapeutic options.

Beyond chronic pain, Gestala is simultaneously exploring areas such as psychiatric disorders, stroke rehabilitation, and Alzheimer’s disease. Peng Lei believes that these indications share common characteristics: a large patient population, strong demand for non-surgical interventions, and greater suitability for identifying new therapeutic pathways from the perspectives of brain state recognition and neuromodulation.

The Imagination of the Ultrasound Pathway Lies in the Possibility of "Holistic Read-Write"

Why Choose Ultrasound? The first answer given by Peng Lei is that it holds the promise of enabling more holistic “reading and writing” of the brain.

Traditional electrical brain-computer interfaces are better at reading high-precision electrical signals from localized areas. For example, to achieve motor control, it is often necessary to obtain clear signals near the motor cortex. However, the brain is not simply composed of isolated regions; complex functions such as cognition, emotion, pain, and memory often involve coordination among multiple brain regions and networks.

“The biggest challenge with electrical methods is that you cannot implant electrodes throughout the entire brain, from left to right and from the surface to deep structures. Electrical techniques excel at decoding local signals, whereas ultrasound offers the potential to capture the whole picture,” said Peng Lei. The company name “Gestala” also derives from the concept that “the whole is greater than the sum of its parts.” “We believe that the brain, too, is an entity where the whole exceeds the sum of its parts.”

From a technical perspective, ultrasound can be used not only for neuromodulation but also holds promise for understanding brain activity through changes in signals such as blood flow. Peng Lei introduced that Gestala focuses not merely on single-channel EEG signals, but on multimodal brain data, including microvascular blood flow signals, electrical signals, and CT imaging. Compared to single-modal signals, multimodal data are more complex and impose higher demands on algorithms and AI.

This also means that ultrasonic brain-computer interfaces are not without challenges. Peng Lei candidly admitted that the so-called “whole-brain read-write” capability remains a possibility requiring continuous breakthroughs, with a long road ahead before mature applications can be realized. In particular, achieving stable and clear signal acquisition through the skull is itself an engineering and algorithmic challenge.

“But in his view, precisely because the ultrasound approach deals with more complex data structures, its integration with AI is even tighter. ‘Our integration with AI will be one of the core values,’ said Peng Lei.”

Brain-Computer Interfaces: Not a Unidirectional Dependence on AI, but Mutual Enhancement

Amid the rapid advancement of large AI models, the relationship between brain-computer interfaces (BCIs) and AI is being reexamined. Previously, AI was primarily regarded as a tool for decoding brain signals; now, neuroscience itself may, in turn, inspire AI development.

Peng Lei cited as an example that there is a complex relationship between neuronal firing and hemodynamic changes. In some brain regions, neural activity may precede changes in blood supply; differences also exist among different types of neurons. To understand these complex phenomena, traditional signal processing methods alone are insufficient; AI models are required to model multimodal, large-scale, and longitudinal data.

“Brain-computer interfaces do not rely on AI; rather, they mutually reinforce each other,” stated Peng Lei. On one hand, AI can help brain-computer interfaces better understand and decode brain activity; on the other hand, neuroscience research into the working mechanisms of the brain may also provide inspiration for next-generation AI.

He believes that for AI to continue advancing, it must address challenges such as continual learning, hierarchical memory, and long-term reasoning—issues that are highly relevant to neuroscience. The human brain remains the only known general intelligence system validated by natural evolution. A true understanding of the brain could not only provide new therapeutic approaches for neurological disorders but also offer new theoretical foundations for the development of artificial intelligence.

However, Peng Lei also emphasized that more data is not always better for AI training; data quality is more critical. One important reason Gestala chose to focus on serious medical scenarios first is that clinical data features clearer annotations, well-defined indications, and more rigorous collection protocols.

“Why did we prioritize serious medical applications? On one hand, it allows us to acquire multimodal data, such as MRI, CT, Doppler ultrasound, and electrophysiological data; on the other hand, it focuses on specific diseases with well-defined patient cohorts and clearly annotated labels,” said Peng Lei. He emphasized that high-quality clinical data will serve as the foundation for subsequent model iteration and product generalization.

“From Complex to Simple”: Transitioning from Serious Medical Devices to Home Healthcare

Regarding the product roadmap, Peng Lei’s judgment is relatively clear: ultrasound-based brain-computer interfaces should not be developed as consumer electronics from the outset; instead, development should begin with serious medical devices.

“Our path must begin with developing serious medical devices, completing validation of safety and efficacy, then expanding to different indications, accumulating more data, and refining algorithms and models,” said Peng Lei. Only after thorough validation in hospital settings can products gradually evolve from Class III serious medical devices to home-use medical devices, and further extend to broader consumer-grade applications.

In his view, the transition of products from hospitals to homes does not mean they shed their medical attributes. Even if future devices become more portable and are used in lighter scenarios such as sleep regulation, their premise must still be based on the safety, efficacy, and data closed-loop required under the medical device pathway.

Peng Lei stated that brain-computer interfaces involve the human nervous system and cannot be driven solely by conceptual packaging and consumer-grade experiences. Validating the technology in serious medical scenarios first, although a more challenging and lengthier path, is more conducive to establishing technical barriers and industrial trust.

Gestala plans to launch its first-generation product by the end of this year and disclose progress in clinical data, scientific publications, and AI models. Subsequently, the company intends to spend approximately 18 months advancing large-scale clinical trials and regulatory submissions. Peng Lei stated that he hopes to make progress in obtaining medical device certification for ultrasound brain-computer interface technologies in the second half of next year.

Editor: Ge Jiaming