Brain Computer Interfaces: Where Neuroscience Meets Technology

By Medha Vemuri

Edited by Akeesha Subasinghe

Abstract 

This paper looks at how a brain computer interface works, including signal processing and feature translation for insights into improving the quality of life of paralytic patients. Furthermore, significant limitations, ethical issues and notable examples of investment into the technology are discussed as well.

Introduction 

Brain computer interfaces (BCIs) are complementary systems that provide an alternative avenue of movement for paralytic patients. Specifically, they enable the patient to interact with their surroundings purely through their ‘thoughts’, with no physical movement or muscular interaction. For example, the systems can enable a patient to move a prosthetic limb, exoskeleton or a wheelchair, thereby improving the quality of life of the patient where the technology has immense potential to influence patient care; cutting-edge research into this is making long strides in that direction.

To understand the scale and potential for impact of these systems, it is first necessary to understand how they work. Simply put, BCIs acquire brain signals, analyse them and translate them into commands that are relayed to output devices like a prosthetic limb or an exoskeleton. The BCI system can be decomposed into a set of components: 

• A device that collects/acquires brain signals - typically micro-electrodes placed either intracranially or on the scalp of the patient.

• A software module/program that clears background static noise in acquired signals and then analyses them. This software module runs/executes on a dedicated hardware processing unit (which typically sits inside a specialised on-body controller), which connects with the micro-electrodes wirelessly.

• A decoding algorithm which translates the analysed signals from the previous stage into commands which are relayed to the output device. This algorithm, again, runs/executes on the dedicated hardware processing unit sitting inside the on-body controller. 

• An output device (like a spelling program on a computer/exoskeleton). The exoskeleton here consists of actuation systems (to generate muscle movement) and a control system which executes the commands relayed by the BCI.

Although BCIs are still in clinical trials, many large companies around the world are choosing to invest in this technology; a notable example of which is Neuralink, who recently started clinical trials in Great Britain. (1)

 How does the BCI system work?

A BCI system operates in three key stages - signal acquisition, feature extraction through analysis, and feature translation. The output of the feature translation stage are the commands which control the output device.

Signal acquisition 

Signal acquisition is the measurement of brain signals/activity using a particular sensor modality by placing a scalp or intracranial electrode on the patient's brain. (2)

There are two main ways of collecting said brain signals: either through invasive procedures or through non-invasive methods. 

1. Invasive approaches 

Invasive approaches of brain signal acquisition involve placing penetrating micro electrodes directly in or on the brain, (typically within the gray matter of the cerebral cortex) to detect and record the electrical impulses produced by neurons, which are collected in their continuous wave form as analog signals (3 , 4). Doing so provides the best, most accurate conversion of neural brain signals to device action (5). This is primarily because the gray matter has an abundance of neuronal cell bodies, therefore allowing for the capture of individual neuron activity with high precision. However, it is important to note that there are significant disadvantages of this method of signal acquisition as it entails high short-term risks like hemorrhage, cerebral edema and general anaesthetic risks. Furthermore, as the micro-electrode is recognised as foreign by the body, it can lead to inflammatory responses by the macrophages and lymphocytes, and in the long term could also lead to the formation of scar tissue around the micro electrode. Additionally, a significant disadvantage is the sheer cost of the surgery (ranging from $40,000 - $100,000). The high cost of the surgery already excludes a significant portion of the 15 million people around the world living with paralysis from even considering invasive procedures as an option, thereby posing questions on the accessibility and reach of the technology.

2. Non-invasive approaches

Alternatively, non-invasive BCI signal acquisition hardware provides a safer, more accessible option that can be used in research and industry on a wider scale. This method of signal acquisition often involves using external tests/ sensors that are placed on the scalp to detect and collect brain signals. More specifically, this process is done through EEG testing. Electroencophelography (EEG) is a non-invasive test which records the brain's electrical activity through sensors which are placed on the scalp. The machine represents the micro voltage fluctuations in the brain using a system of electrodes, amplifiers and a computer which detects the signal, magnifies it and displays it on said computer. 

Another method of non-invasive signal collection is through the use of fNRIS machines. fNRIS machines or (function Near Infrared Spectroscopy machines) involves estimating the changes in optical Blood Oxygen Level Dependent Signals (BOLDS) - an indicator of neural activity, induced by the changes in absorption of near infrared light; instead of measuring electrical signals (6). A head cap containing micro LEDs is placed on the patient’s head and infrared light is shone onto it. To record the required data, this system capitalises on the fact that infrared radiation is absorbed differently by tissues and blood. More precisely, the pulsatile nature of arterial blood flow is the key to how this system works. With each systole of the heart beat, there is a temporary physical expansion of the tissue, leading to corresponding changes in the amount of infrared radiation absorbed or reflected. With each diastole of the heart beat, the tissue relaxes and the amount of infrared radiation changes back. 

Feature extraction 

Feature extraction is the process of analysing the brain signals collected from the previous stage to differentiate signal characteristics, representing them in binary data format for translation into output commands. This process begins by firstly removing background static and other noise which is inadvertently picked up by the electrodes during the brain signal acquisition stage through basic digital signal processing techniques, before translation.

To improve the efficacy of this process for both the patient and the system, there is a BCI ‘training period’ where the user and BCI system ‘work together’ to personalise the system to the user’s needs and brain signals. During this process, the patient performs a series of mental tasks such as imagining a certain movement, which is followed by the electrode recording corresponding brain activity(10). This data is used for the program to understand and recognise thought processes by recognising patterns in brain activity. 

The time it takes for this process ranges anywhere from hours to months. Furthermore, different factors like the type of signal acquisition hardware involved, the different methods of calibration, prior experience of the patient with BCIs and the complexity of the BCI task can greatly influence the learning curve.

Since the first introduction of this technology in the early 2010s, research into BCI systems has come a long way. The development of machine learning (ML) algorithms since then has played a key factor in improving the efficacy of the system as a whole. The impact of this can be clearly seen in BCI system training times, which used to span from several hours to even months in its early days, but has now been significantly cut down to a span of hours (7). 

Ultimately, training the BCI system is mainly about personalisation of the algorithm to tailor it to individual neural patterns by building a ‘shared language’ between the brain and machine. 

Feature translation and machine learning

Feature translation is the crucial step that maps extracted, meaningful brain signal patterns into output commands such as ‘move cursor up’ or ‘move hand left’.

Here the extracted features are fed into a machine learning model, to quickly and easily classify the difference in commands. Machine learning (ML) is the field of the study of programs or systems that train models to make predictions from input data. ML powers important everyday technology, like google maps and translation apps. As a simple analogy, a model is trained using a set of labeled images. In this example 1000 images labeled ‘cat' and 1000 labeled ‘dog’ are inputted into the model. The model analyses numerical data in the image, like edges, shapes, textures. On image 2001, the model uses the patterns it learnt to predict whether the image is more likely to be a cat or a dog.

When AI works within BCI signal translation, raw neural signals such as signal pulse duration, amplitude or other electrical properties of the neural tissue are provided to the model constantly. The AI model (which runs on an external computer) learns internal parameters through training on that data, and classifies them into commands corresponding to brain signal patterns like ‘yes’, ‘no’, or ‘move right’. A crucial consideration that must be understood here is that the AI model must be dynamic, as brain signals are dynamic and change based on age and hormone levels. Hence it is vital that the AI model is adaptable and dynamic to accommodate the neuroplasticity of the patient.

Output device

The commands from the feature translation stage are then relayed to the external device providing functions such as selection, cursor control, robotic arm operation and so forth. 

Limitations and the ethics behind it all 

One of the key issues that can impact the development of these systems is the lack of specific standards that govern the development of these BCI systems. The lack of standards leads to siloed innovations which are not interoperable.

Also as mentioned by Qiu L, Tang S, Meng J, et al(8), BCI applications can have unrestricted access to brain signals where sensitive information can be accessed without the patient's knowledge, posing questions on the ethics of the system- questions which are especially vital today as more companies around the world invest in BCI technology. 

As mentioned before, the cost of the BCI system proves to be a vital issue surrounding the technology today, posing particularly vital implications about the accessibility of the system to the wider public. 

Notable mentions 

A US Based neuro tech company, Synchron is a pioneer in the research into BCI technology, making use of Apple’s newly released BCI HID to transmit signals from the paralytic patient to an iphone/ipad wirelessly via bluetooth. More importantly, instead of invasive or even non-invasive signal acquisition hardware, they use the “The Stentrode™ Platform" in their BCI systems. This device mimics a cardiac stent (but with a circuit board) and is delivered via a catheter to a main vein in the brain called the superior sagittal sinus (SSS) . The SSS is positioned directly next to the motor complex in the brain. The close proximity between the vein and the motor complex allows the Stentrode™ Platform to precisely and accurately collect brain signals. Furthermore, as this is placed endovascularly, it significantly reduces the risks associated with open brain surgery, proving to be the safest option in the market today. (9). 

Conclusion

In conclusion, brain computer interface systems irrevocably prove to be extremely beneficial to health and wellbeing, offering potential for the improvement of the quality of life of a significant portion of the global population, despite the risks associated with it. According to the WHO, about 15 million people around the world live with paralysis and spinal cord injuries today. Owing to the commonality of the condition, the potential for impact of this technology is immense, helping paralytic patients regain control over their lives, not only providing physical benefits but also improving their mental health. To compound this, although this technology is currently in clinical trials,  increased investment into this technology only proves that it is not too far from mainstream adoption, thereby being able to become more accessible to the wider public. 

Bibliography

1. Neuralink. 2025. “Clinical Trials | Neuralink.” Neuralink. 2025. https://neuralink.com/trials/.

2. Shih, Jerry J., Dean J. Krusienski, and Jonathan R. Wolpaw. 2012. “Brain-Computer Interfaces in Medicine.” Mayo Clinic Proceedings 87 (3): 268–79. https://doi.org/10.1016/j.mayocp.2011.12.008.

3. Toews, Rob. 2025. “The Past, Present and Future of Brain-Computer Interfaces.” Forbes, October 5, 2025. https://www.forbes.com/sites/robtoews/2025/10/05/these-are-the-startups-merging-your-brain-with-ai/.

4. Cleveland Clinic. 2023. “Grey Matter.” Cleveland Clinic. Cleveland Clinic. March 19, 2023. https://my.clevelandclinic.org/health/body/24831-grey-matter.

5. Obien, Marie Engelene J., Kosmas Deligkaris, Torsten Bullmann, Douglas J. Bakkum, and Urs Frey. 2015. “Revealing Neuronal Function through Microelectrode Array Recordings.” Frontiers in Neuroscience 8 (January). https://doi.org/10.3389/fnins.2014.00423.

6. “Functional Near-Infrared Spectroscopy.” 2020. Wikipedia. December 10, 2020. https://en.wikipedia.org/wiki/Functional_near-infrared_spectroscopy.

7. Kawala-Sterniuk, Aleksandra, Natalia Browarska, Amir Al-Bakri, Mariusz Pelc, Jaroslaw Zygarlicki, Michaela Sidikova, Radek Martinek, and Edward Jacek Gorzelanczyk. 2021. “Summary of over Fifty Years with Brain-Computer Interfaces—a Review.” Brain Sciences 11 (1): 43. https://doi.org/10.3390/brainsci11010043.

8. Sun, Xiaoyu, and Bin Ye. 2023. “The Functional Differentiation of Brain–Computer Interfaces (BCIs) and Its Ethical Implications.” Humanities and Social Sciences Communications 10 (1). https://doi.org/10.1057/s41599-023-02419-x.

9. “Home.” n.d. Synchron. https://synchron.com/.

10. Yu, Hongli, Sidi Ba, Yuxue Guo, Lei Guo, and Guizhi Xu. 2022. “Effects of Motor Imagery Tasks on Brain Functional Networks Based on EEG Mu/Beta Rhythm.” Brain Sciences 12 (2): 194. https://doi.org/10.3390/brainsci12020194.