Brain-Computer Interfaces in Modern Medicine
Written By Mubina Shahnaz¹ and Elgin Tawiah²
Illustrated By Aarna Mishra¹
Edited by Jeffrey Bartes²
1. National University of Singapore Medtech
2. University of Pennsylvania Grey Matters
Introduction
Imagine a world where a person paralyzed from the neck down can navigate a smartphone, typing out a message without lifting a finger… just using their thoughts. What sounds like the opening scene of a sci-fi novel is actually turning into reality in our medically advanced society. Recent breakthroughs in Brain-Computer Interfaces (BCIs) are building an explicit informational highway between human gray matter and digital machinery. With the evolution of BCIs, the potential to diagnose, treat, and manage some of our most devastating neurological and motor disorders is limitless. BCIs are used to create highly engaging, personalized cognitive training protocols via integration with other up-and-coming technologies like VR. These promote neuroplasticity and slow down executive decay. Read more to find out how BCIs are transitioning from clinical trial laboratories to everyday medical devices.
Clinical Applications
Brain-computer interfaces can improve clinical outcomes for debilitating conditions. An instance of this includes stroke. Stroke causes hemiparesis, paralysis typically present on one side of the body. Unfortunately, as a result, this causes gait abnormalities and reduced strength, making daily living activities such as eating and brushing teeth difficult.
Attempts to address this issue in the past included stroke rehabilitation, which resulted in motor improvements plateauing just three months after a stroke, still resulting in permanent motor deficits.
This is why the necessity to develop a device or mechanism to assist stroke patients in order to improve the feasibility of daily tasks was urgent. During the mid 2000s, Dr. Eric Leuthardt and Dr. Daniel Moran developed the IpsiHand Rehabilitation System (IpsiHand System), the first BCI therapy for stroke patients. The device uses an electroencephalogram (EEG), a non-invasive device that measures brain waves, to record signals from the unaffected side and control a wearable exoskeleton around the affected hand (Rustamov et al., 2025). Evidence suggests the facilitation of motor recovery in chronic stroke patients.
Since then, continual improvements have been made to BCI systems for stroke patients, such as the incorporation of upper extremities and the use of real-time feedback by the BCIs, allowing patients to adjust their learning. This feedback system promotes motor learning and neuroplasticity, the brain’s ability to change and to form stronger neural connections. This added neuroplasticity is key in regaining lost function while optimizing existing functions in the brain (Tonin et al., 2025).
Beyond motor restoration, BCIs can assist patients suffering from neurodegenerative conditions like Amyotrophic Lateral Sclerosis (ALS) and Alzheimer’s Disease.
ALS destroys motor neurons throughout the body, even those responsible for digestion and speech. Stephen Hawking was famously known to suffer from early-onset ALS at the age of 21. Though ALS imposed several challenges to his daily life, one important aspect of his life was not lost: communication. His ability to communicate his understanding of the universe and bridge the gap between general relativity and quantum mechanics.
BCIs allow speech to be restored. EEGs can be used to capture brain signals when ALS patients try to articulate words, which are then processed by machine learning algorithms to identify the word that is being attempted to be spoken. This is later translated into speech output by a device (Deng et al., 2025).
Then, Alzheimer’s disease poses a challenge to BCIs since BCIs are based on instrumental learning and require users to self-regulate their thoughts, which is largely absent in AD. However, advancements are still in the process of being made (Liberati et al., 2012).
Algorithmic Transformation of Neural Data
So how does algorithmic transformation of neural data work? When scientists first started out, they relied heavily on mathematical formulae to identify and rationalize motor cortex activity. Soon, they discovered the serious flaws of that system. The human brain is highly dynamic and non-linear. The same EEG waves could be attributed to a multitude of factors like the patient’s fatigue level, attention span just to name a few. Although our foundations are strongly built upon these static frameworks, we have shifted towards adaptive data-driven frameworks now. Modern BCI software does not force the brain to fit into a rigid mathematical box. Instead, the software architecture adapts to the user, treating neural decoding as an ongoing, fluid conversation that evolves with the user’s real-time brain state.
To further enhance the uses of BCIs, the system can also be coupled with deep learning models. Deep learning models are forms of artificial intelligence that are akin to the human brain. In a similar sense, they are composed of “neurons” that process input data like brain waves and convert them into commands such as moving a hand. The brain waves that are into deep learning models are entirely dependent upon the user’s actions. Instead of neurons, deep learning models work on neural networks, which are functionally similar. Neural networks receive inputs at input nodes that are weighted, meaning certain inputs to the model are deemed more valuable in terms of determining what the output response will be. Going from inputs to outputs.
As expected, the neural network does not get the correct output on the first attempt. In fact, neural networks require a lot of data to get an accurate response. There are many training rules adapted to ensure that an accurate response is achieved, with an example being the delta rule.
Get Nusmedtech Official’s stories in your inbox
Join Medium for free to get updates from this writer.
The delta rule first begins by assigning weights, which is the importance or strength of an input. Then, over time, these weights are adjusted accordingly as inputs are fed into the neural network until a correct output is obtained. It is important to note that there are other training models aside from the delta rule with hidden layers between input and output nodes, increasing the complexity of training a deep learning model.
Neural networks also make use of backpropagation to further training accuracy, which is essentially the reverse of the training rule. Backpropagation starts from the output node and works backwards to input nodes, improving both computational efficiency and learning accuracy.
In all, combining the training rule and backpropagation allows for increased accuracy in BCI by taking raw, messy brain waves and encoding them into actionable commands (Mansoor et al., 2020).
Two of the most fatal bottlenecks have been real-time noise filtering and automated classification. Both are being effectively tackled by robust deep learning models. Traditional filters often screen away essential data together with noise. To reduce the propensity of misclassification of vital signals as artifacts, Convolutional Neural Networks (CNNs) are trained thoroughly. Through multi-layered pattern recognition, CNNs are able to clean the data with better accuracy for further processing. Next, machine learning models are adept at distinguishing subtle signals from specific thoughts. With the help of Recurrent Neural Networks (RNNs), BCIs predict a user’s intended movement or continuous speech a fraction of a second before it is fully formed.
Analysis of Human Clinical Trials Timeline
Since the early 2000s, neurotechnology has transitioned from sole computer simulations to actual implantable interfaces. Tracing the history of BCIs from bench side to market highlights a steep trajectory of innovation. It also reveals a consistent roadblock we have faced, that is the critical lack of standardization across a fragmented scientific landscape. 1998 was when the first stroke patient had the long-term neurotrophic electrode implanted. It was pivotal in establishing that wireless brain signals could be utilized for a patient to navigate a computer cursor. Following this, in 2006, multi-electrode arrays entered the scene. It enabledpatients with spinal cord injuries to operate robotic hands and handle 2D cursor spaces. This evolved into self-feeding robotic arm control by 2012.
A decade later, a study in 2021 demonstrated that patients can go beyond pointing at letters using computer cursors. Patients could output text of 18 words per minute just by visualizing their hand movements for the letters in their brain. This was groundbreaking in extracting language directly from the mind. Just 2 years later, intracortical microelectrode arrays and electrocorticography (ECoG) proved that unvoiced thoughts can also be output into text at an unprecedented speed of up to 79 words per minute. With that research shifted to areas exploring the lifespan of these implants. Signal decay caused by micro-motions and local tissue reactions. Regular silicon-based microelectrode arrays can last up to 8.5 years while specialized neurotrophic glass-cone electrodes have shown impressive performance for 1.5X of the regular ones.
Despite these innovations, a persistent challenge that we face is the lack of structured, open-science data-sharing models. While some levels of non-transparency are critical to protecting patient privacy, developing non-standardized outcome assessment metrics to evaluate performance is equally important. Without standardized benchmarking and collaborative open science, it is difficult to objectively compare signal-processing algorithms, hardware stability, or user experiences across different trials.
Conclusion
Brain Computer Interfaces are, simply put, promising. They are rapidly transitioning from lab spaces to clinical trials. Although some studies of the applications of artificial intelligence (AI) to BCIs lack reproducibility and generalizability, the growth of artificial intelligence has promising results. For instance, the processes of analysis and decoding neural activity has expeditiously increased through AI, expanding our capabilities of assisting paralyzed patients and neurophysiological discoveries (Zhang et al., 2020).
Furthermore, estimates go so far as to claim that by the year 2030, the BCI market is expected to reach $6 billion and roll out medically and commercially. Through increased efforts, movement can be restored, speech can be restored, sight can be restored, and so many other facets that were previously either thought to be impossible or implausible can be achieved. BCIs create a promising future (Strategic Intelligence, 2025).
Sources
Deng, Q., Fu, Z., Ma, N., & Wang, B. (2025). Application and future directions of brain-computer interfaces in neurological disorders: Technological advances, clinical practices, and challenges. Brain Hemorrhages. https://doi.org/10.1016/j.hest.2025.09.002
Liberati, G., Dalboni da Rocha, J. L., van der Heiden, L., Raffone, A., Birbaumer, N., Olivetti Belardinelli, M., & Sitaram, R. (2012). Toward a brain-computer interface for Alzheimer’s disease patients by combining classical conditioning and brain state classification. Journal of Alzheimer’s Disease: JAD, 31(Suppl 3), S211–S220. https://doi.org/10.3233/JAD-2012-112129
Mansoor, A., Usman, M. W., Jamil, N., & Naeem, M. A. (2020). Deep learning algorithm for brain-computer interface. Scientific Programming, 2020, 1–12. https://doi.org/10.1155/2020/5762149
Rustamov, N., Souders, L., Sheehan, L., Carter, A., & Leuthardt, E. C. (2025). IpsiHand brain-computer interface therapy induces broad upper extremity motor rehabilitation in chronic stroke. Neurorehabilitation and Neural Repair, 39(1), 74–86. https://doi.org/10.1177/15459683241287731
Strategic Intelligence. (2025, August 13). Brain-computer interfaces are closer than you think. Clinical Trials Arena. https://www.clinicaltrialsarena.com/analyst-comment/brain-computer-interfaces-closer/
Tonin, A., Semprini, M., Kiper, P., & Mantini, D. (2025). Brain-computer interfaces for stroke motor rehabilitation. Bioengineering (Basel, Switzerland), 12(8), 820. https://doi.org/10.3390/bioengineering12080820
Zhang, X., Ma, Z., Zheng, H., Li, T., Chen, K., Wang, X., Liu, C., Xu, L., Wu, X., Lin, D., & Lin, H. (2020). The combination of brain-computer interfaces and artificial intelligence: Applications and challenges. Annals of Translational Medicine, 8(11), 712. https://doi.org/10.21037/atm.2019.11.109
