Brain-Computer Interfaces: From Science Fiction to Reality
At present, the brain-computer interface has passed the technical demonstration period and is about to enter the period of large-scale industrialization. Brain-computer interface is an interdisciplinary discipline in the three fields of neuromedicine, electronic technology and artificial intelligence, and related enterprises need to have strong technical accumulation in three directions. Overall, the feasibility of invasive brain-computer interface is high; At present, after a series of technical iterations of invasive brain-computer interfaces, the core pain points of craniotomy such as high threshold and short product life have been basically solved. In the brain-computer interface industry chain, Neuralink's technical solutions have the most commercial prospects. In May 2023, the U.S. FDA approved Neuralink to conduct human clinical trials; At the end of January 2024, Musk confirmed that Neuralink had completed its first human implant and was in good condition. With the verification of Neuralink's technology path and business model, invasive brain-computer interfaces are expected to be implemented on a large scale in medical scenarios first.
Brain-computer interface (BCI) technology enables the exchange of information between the human brain and a computer. Brain-computer interface is the process of human consciousness and computer interaction by collecting brain nerve signals, which usually includes two parts: neural recording (reading) and neural regulation (writing): 1) neural recording (reading): electrodes collect electrical signals from brain neurons, and extract features and translate them through artificial intelligence and other technologies, transforming bioelectric signals into information that can be understood by computers; 2) Neuromodulation (writing): If you want to realize the control of the human brain and limbs, you also need to implant electrodes in the limb nerves, and transmit computer instructions to neurons such as the brain and limbs through the electrodes to realize the control of the human body. Together, neural recording (reading) and neuromodulation (writing) form a closed-loop system of brain-machine-brain. At present, brain-computer connection mainly solves the problem of data transmission from brain to computer, and is expected to be implemented on a large scale in the near future. The computer-to-brain data transmission is still in the stage of breakthrough in the underlying technology.
Brain-computer articulation has the potential to revolutionize the paradigm of human information transmission. In the scenario where people export information to the outside world, the common way of information transmission is that people express their opinions through words or words (100-200 words per minute); In the context of receiving information from the outside world, sound, text, and vision must be used as media. In contrast, brain-computer connection** greatly increases the bandwidth of information transmission. When people transmit or receive information through brain-computer interfaces, brain-computer interfaces directly use consciousness as the medium of communication, replacing low-bandwidth language and text media, which greatly improves information density.
Brain-computer articulation can be evaluated in terms of performance and usability. Performance metrics include response time, recognition accuracy, number of outputable instructions, and Fitz throughput; Usability metrics include ease of use, longevity, robustness, security, and interoperability. Ease of use can also be seen in terms of preparation, length, lightness and comfort.
Brain-computer interfaces require strong signal acquisition performance and data processing capabilities. The nine subdivisions of brain-computer interface can ultimately be boiled down to signal acquisition capabilities and data processing capabilities. Since it cannot be directly connected to the nervous system, the electrodes of the brain-computer interface mainly receive signals through the waves stirred up around the nervous system electrical signals during the propagation process, and the signal intensity is much smaller than that inside the nerve, so the signal acquisition ability of the attached electrodes is required. In addition, electrodes cannot directly understand the nervous system, and can only collect a large amount of data to summarize the nervous system in different states such as excitement, relaxation, and sleep, so the accuracy of neuronal electrical signal analysis is required.
Key competitive element 1: signal acquisition capability The signal acquisition technology path of brain-computer interface can be divided into two categories: invasive and non-invasive; Both invasive and non-invasive methods essentially detect brain activity by collecting neuronal electrical signals, but the difference is that the electrodes are mounted and the intensity of the collected neural signals is different. At present, the common invasive techniques include technical electrocortical (ECOG), local field potential (LFP), single neuron action potential (SPIKE), etc. Non-invasive techniques include electroencephalography (EGG).
Intrusive signal acquisition is stronger, while non-invasive signal acquisition is weaker. The selection of the signal source is the starting point of the brain-computer interface system design, and it also basically determines the technical scheme and engineering route of the corresponding system, as well as the system performance (such as the degree of freedom of motion control, accuracy, system delay, etc.) and application scenarios that can be realized. Invasive can collect stronger neural voltages and signal frequencies because they are closer to the neurons.
Technical route: invasive brain-computer interface
Invasive brain-computer interfaces collect and record signals from tissues deep below the skull, requiring a craniotomy on the user. The invasive type has a high degree of compatibility with neurons and has the strongest signal acquisition ability; However, due to the technical difficulty and cost of use, the current audience is small, and the downstream is mostly medical scenarios. Faced with the pain points of invasive brain-computer interfaces requiring craniotomy and rejection reactions, various manufacturers have adopted flexible and miniaturized surgical robots and electrodes to deal with them. At present, the leading manufacturers of invasive brain-computer interfaces include Musk's Neuralink, Blackrock Neurotech, Braingate, etc.
Invasive brain-computer interface is a concept of brain-computer interface in a narrow sense, and it is also the most promising application direction. Invasive is the one with the greatest practical benefits among the three technical paths of brain-computer interface. Compared with non-invasive brain-computer interfaces, invasive electrodes are closer to the nervous system, so the recorded signals have high spatiotemporal resolution, large amount of information, and can be controlled in real time and accurately for complex tasks, so they perform well in four indicators: response time, recognition accuracy, number of outputable instructions and Fitz throughput. There have been dozens of successful cases of invasive brain-computer interfaces, which have been shown to provide people with disabilities with additional mobility and communication skills.
At present, the main bottlenecks of invasive brain-computer interfaces are the difficulty of craniotomy and the short service life
1) Craniotomy is required to install implantable brain-computer interface devices, but craniotomy is technically difficult and difficult to be replicated in commercial batches. Taking the United States as an example, there are currently more than 5 million patients with varying degrees of paralysis in the United States, while there are only more than 150 neurosurgeons in the United States who are qualified for craniotomy.
2) Under the current state of technology, it is difficult to guarantee the surgical prognosis of implanted brain-computer interface. The electrodes inserted into the brain produce a rejection reaction, turning the brain tissue near the electrodes into scar tissue that covers the electrodes and blocks the work. The current commonly used Utah array electrodes have a lifespan of only 2-5 years, and implanters will need to undergo brain surgery again if they do not want to return to a disabling state.
The surgical robot can complete the whole process of craniotomy, alleviating the pain points of the current craniotomy that is difficult and the supply is scarce. For example, Neuralink's surgical robot R1 first determines the location of the implanted chip through methods such as functional nuclear magnetic resonance. After the position is determined, the surgical robot will make an incision on the scalp, use a special hole opener to open a round hole in the skull of the same size as the chip, and then peel off the underlying dura mater to expose the cerebral cortex tissue. With the assistance of an optical system, the surgical robot inserts 64 wires on the electrodes into the cerebral cortex, each wire contains 16 potentials, and a total of 1024 potentials of the 64 wires will receive electrical signals near the neurons and transmit them to the chip for preliminary analysis.
Technical route: non-invasive brain-computer interface
Non-invasive brain-computer interfaces do not require an opening in the skull, but instead attach electrodes to the surface of the scalp or near the scalp to collect brain response signals, and obtain nervous system information by means of electroencephalography and MRI. However, because they are separated from the skull, the electrodes of the non-invasive brain-computer interface cannot listen to specific clusters of neurons, and can only receive the sum current of scattered electrical signals collected by the whole brain, and because they are far away from the intracranial nerves, the collected signals often contain a lot of noise. Therefore, the information that can be obtained by non-invasive brain-computer interfaces is relatively limited, and it can generally only interpret the overall state of the brain, such as the degree of wakefulness, emotions, etc., and it is difficult to accurately obtain specific intentions, perceptions and other information.
Non-invasive BCIs have a low barrier to entry.
Compared with the invasive type, the non-invasive brain-computer interface system has higher safety and extensiveness, and the types of signals that can be used are more abundant, forming a brain-computer interface system based on EEG, encephalography, functional near-infrared and functional magnetic resonance imaging. Among them, the brain-computer interface system based on EEG has received extensive attention due to its advantages in terms of cost and portability, and has become the main focus of non-invasive brain-computer interface.
At present, the performance and usability of non-invasive brain-computer interfaces are still insufficient. As mentioned above, one of the core competitive elements of brain-computer interfaces is the signal acquisition ability, while non-invasive brain-computer interfaces have low spatial resolution because the sensing electrodes are far away from the nerves, and the obtained signals are doped with a lot of noise, and because they can only detect the whole brain nerve signal, the spatial resolution is low. Therefore, the performance indicators and usability indicators of non-invasive brain-computer interfaces are lacking, and there is still a long way to go before large-scale implementation. Non-implantable brain-computer interface is mainly used in training, education and entertainment, intelligent life, manufacturing and other scenarios. In the field of non-implantable brain-computer interface, the industry generally attaches great importance to the research oriented to the industrial and consumer fields, and cooperates with peripherals such as virtual reality, augmented reality, eye tracker, exoskeleton, etc., and uses non-implantable brain-computer interface systems to carry out multi-scenario application exploration, such as: sleep state monitoring, sports training, using user brainwaves to create and control electrical appliances, and using user emotion recognition data to recommend users personally.
Key Competitive Factor 2: Data Processing Capability
Data processing is a key technology downstream of brain-computer interface, which directly determines the recognition accuracy of brain-computer system. Brain-computer interface data processing can be divided into three capabilities: data cleaning, feature extraction, and data analysis
Data cleaning: Ensuring the quality, accuracy, and credibility of neural signal data has a significant impact on the subsequent analysis, modeling, and decision-making process. The cleaning process is not only about removing noise and reducing the time and effort of dealing with anomalies and errors during subsequent analysis, but also about preserving feature information as much as possible.
Feature extraction: Select, transform, and annotate new features from the cleansed data to better represent key information about the data. Feature extraction can reduce the data dimension and select the most representative features to reduce the complexity of the data and improve the computational efficiency and model performance. At the same time, it reduces the computational resources and time required to train the model, so that the model converges faster and improves the training efficiency. In addition, feature extraction can also improve the interpretability of the model and reduce the complexity of the model. Feature extraction can identify and label the frequency and shape of neural peaks, which is convenient for deep learning models to determine.
Data analysis: At present, the data analysis of brain-computer interfaces is mainly realized through deep learning models. In the case of Neuralink, there are 1024 potentials on the N1 electrode, which generates a huge amount of data every second, and only a deep learning model can analyze it in a timely and accurate manner. At present, the deep learning models used in the field of brain-computer interface are mainly convolutional neural networks (CNNs) for complex feature extraction and recurrent neural networks (RNNs) for longer time series data. In addition, in addition to the training of the model itself, the accumulation of a large number of training datasets also plays a crucial role in improving the accuracy of deep learning models.
The market prospect is broad, and star companies are blooming
Favorable policies continue to be released to guide the growth of the industry. Brain-computer interface is expected to become the main battlefield of the next cross-integration of life science and information technology, countries have launched brain science research plans, China is also actively promoting the development of brain-computer interface industry, and relevant policies and action plans have been released. In 2016, China's "13th Five-Year Plan" listed "brain science and brain-like research" as a "major national scientific and technological innovation and engineering project", marking the full development of the "China Brain Plan". In August 2023, the Ministry of Industry and Information Technology (MIIT) issued the Implementation Plan for the New Industry Standardization Pilot Project (2023-2035), emphasizing the need to promote the standardization of brain-computer interfaces.
There are many potential user groups downstream of brain-computer interfaces, and the market scale is broad. According to the statistics of the China Disabled Persons' Federation, there are 24.72 million people with physical disabilities, nearly 18 million people with visual impairments, and 27.8 million people with hearing disabilities. According to incomplete statistics, the prevalence of Alzheimer's disease in China is 6%, and there are more than 10 million patients with other neurological diseases, and it is growing rapidly with the increase of aging. According to McKinsey**, the global potential market size of brain-computer interfaces in the medical field from 2030 to 2040 is US$40 billion, of which US$15 billion is for serious medical care and US$25 billion for consumer healthcare, with a compound annual growth rate of more than 10%.
From the perspective of industrial chain development, there are currently a large number of non-invasive brain-computer interface companies. As of the first quarter of 2023, there are more than 500 representative brain-computer interface companies in the world. Among them, the upstream accounts for 8%, including enterprises that manufacture and sell electrodes, chips, peripherals, and related core devices; The midstream accounts for 30%, including companies that manufacture and sell medical and scientific tools, analysis software and acquisition equipment: Downstream accounts for 62%, of which 9% are in the invasive technology route and 53% are in the non-invasive technology route. From the perspective of technical routes, most of the enterprises have non-intrusive technical routes, and the low technical threshold is the main reason. Among the more than 500 brain-computer interface-related companies in the world, 20% are engaged in invasive technology research and development, and 80% are engaged in non-invasive technology research and development.
The number of brain-computer interface companies in China is growing rapidly. From a geographical point of view, the United States and China are important countries for brain-computer interface companies. Global brain-computer interface-related companies are active in more than 40 countries, with the number of companies in the United States and China exceeding 100 and being in the first echelon in the world, and the number of companies in Canada, the United Kingdom and Israel being in the second echelon, all of which are more than 20. In 2022, there will be 8 new brain-computer interface companies in China, ranking first in the world.
The financing scale of brain-computer interface enterprises continues to upgrade. Most of the brain-computer interface industry is a first-tier start-up, and the funding mainly depends on venture capital. From 2013 to the third quarter of 2023, there have been nearly 800 global venture capital investments in the field of brain-computer interfaces, with a total amount of more than US$10 billion, and nearly 300 companies have been invested, including angel rounds, seed rounds, and A rounds. From 2019 to 2021, brain-computer interfaces attracted large investments, and the growth rate of investment accelerated, and the total annual investment amount fell after 2022, partly because the brain-computer interface field was dragged down by the industry-wide market investment expectations due to the economic recession, but the financing scale of the industry's leading companies still increased considerably.
At present, the non-medical field occupies a high share of the brain-computer interface market. Based on the different medical application scenarios of brain-computer interface, it can be divided into three categories: serious medical scenarios, consumption scenarios, and serious medical and consumer scenarios. In serious medical scenarios, brain-computer interfaces are used for central nervous system diseases, including organic diseases and functional diseases; In the consumer medical scenario, brain-computer interfaces are used in wearable devices for healthy people, which have the characteristics of low technical barriers, fierce market competition, and a wide audience. In the cross-border scenario of serious medical treatment and consumption, clinically approved products or methods are used for C-end patients, which have the characteristics of high technical barriers and few competitors. In the non-medical field, brain-computer interface applications are also developing rapidly, especially in education, entertainment and gaming industries.
The star companies in the field of invasive brain-computer interfaces are mainly concentrated in the United States. Among them, Musk's Neuralink, Blackrock (Blackrock Neurotech), Braingate, Synchron and other companies have deep technical reserves and rapid commercialization progress; China's inter-brain interface industry is still in the follow-up stage, and the head manufacturers of invasive brain-computer interface include Brain Tiger Technology.
Investment Suggestions Although there is still a certain degree of gap between China's brain-computer interface industry chain and foreign countries, there have been breakthroughs in some links. From a general point of view, the main breakthroughs in the domestic related industry chain are in the upstream EEG acquisition equipment, related algorithms and midstream brain-computer interface related products; In the upstream of the industrial chain, Brain Tiger Technology has realized all the self-development of electrodes, algorithms and operating systems;
In the middle of the industrial chain, domestic manufacturers such as Strong Brain Technology and Macro Intelligence have already landed products, and manufacturers such as Zhentai Intelligence and Yifei Huatong focus on medical scenarios and iterate their products rapidly. Therefore, it is recommended to pay attention to the development of electrodes, algorithms, and operating system related manufacturers in the upstream of brain-computer interfaces, as well as the medical, entertainment, and education-related industrial chains in the midstream.
This is an abridged excerpt from the report, the original PDF of the report
"In-depth Report on Brain-Computer Interface in Information Technology-Software and Service Industry: Passing the Technical Demonstration Period and Entering the Dawn of Industrialization-Yangtze River**[Zong Jianshu]-20240130 [Page 28]".
Report**: Value Catalog