Introduction
Dementia is a progressive neurodegenerative syndrome. The global prevalence of dementia is rising, driven by population ageing and increased life expectancy[
1]. The majority of cases are due to Alzheimer’s disease, while vascular, Lewy body, and frontotemporal dementias account for most of the remaining cases. Significantly, while there has been progress in the understanding of the pathophysiology of the diseases, current treatments are largely symptomatic. The global burden of Alzheimer’s disease has increased substantially. It is projected that there will be an increase in the prevalence and cost of the disease over the coming decades. While there has been extensive research into the pharmacological treatment of Alzheimer’s and other dementias, with amyloid and tau pathologies being the main target, the efficacy of disease-modifying therapies has been limited. This has led to the search for alternative neurotechnologies[
2]. Brain-computer interfaces (BCIs) enable direct communication between the brain and other devices by decoding these signals and using them to command software or hardware[
3]. At their core, BCIs seek to restore, replace, or augment compromised central nervous system functionalities by directly recording and decoding neural activity, delivering potential therapeutic pathways in neurodegenerative disorders[
4]. Although BCIs were initially developed for motor restoration in paralyzed patients, their applications have expanded to include cognitive neuroscience, neuromodulation, and neurodegenerative diseases[
5].
Methods
This narrative review was conducted to synthesize current evidence regarding the application of BCI technologies in dementia and mild cognitive impairment (MCI). A structured literature search was performed across PubMed, Scopus, and Web of Science databases. The final search was conducted in December 2025. The search strategy combined controlled vocabulary and free-text terms, including: (“brain-computer interface” or “BCI”) and (“dementia” or “Alzheimer’s disease” or “mild cognitive impairment”) and (“EEG” or “neurofeedback” or “neuromodulation” or “closed-loop stimulation”). Database-specific adaptations of these terms were applied. The initial search yielded 180 records. After removal of duplicates (n = 35), 145 records were screened based on titles and abstracts. Of these, 112 records were excluded due to irrelevant topics, non-clinical study design, conference abstracts, or non-English language. A total of 33 full-text articles were assessed for eligibility and included in the final narrative synthesis. The study selection process is illustrated in Figure 1. Study selection was based on methodological rigor, clinical relevance, and innovation in BCI architecture. Given the narrative design, formal risk-of-bias scoring and quantitative meta-analysis were not performed. However, a structured qualitative appraisal was conducted. Studies were evaluated based on predefined criteria, including study design (randomized vs. non-randomized), sample size adequacy, presence of control groups, duration of follow-up, standardization of outcome measures, and reproducibility of findings. Greater interpretive weight was assigned to studies with more rigorous methodology, including randomized controlled designs and validated cognitive endpoints. A total of 33 studies were included in the final narrative synthesis. Additional references were used to provide background context and conceptual discussion.
Overview
BCIs have also gained attention in the field of neurosurgery, as they enable the observation, modulation, and implanting of the brain[
6]. In the case of dementia, BCIs show promise for early diagnostic biomarkers, cognitive rehabilitation, improved functional autonomy, and closed-loop treatment methods[
7]. This review combines the present body of evidence on BCI technology in dementia, focusing on neurophysiological mechanisms, clinical applications, and future directions relevant to neurosurgical practice. Although interest in the technology is increasing, clinical application remains limited by small sample sizes, lack of uniform protocols, and insufficient longitudinal evidence. Surmounting these limitations is critical to determine the true therapeutic potential of BCIs in neurodegenerative diseases. Importantly, while the core BCI architecture is generic, its clinical implementation in dementia is application-specific, with distinct adaptations for diagnosis, rehabilitation, and assistive communication Figure 2.
Principles of BCI technology
BCI systems function through four primary stages: signal collection, signal processing, feature extraction, and output translation[
8]. The overall architecture of a closed-loop BCI system in dementia is illustrated in Figure 3. BCIs utilize neural signal acquisition and decoding, which include modalities such as electroencephalography, electrocorticography, and intracortical recordings, which vary depending on the spatial resolution and invasiveness of the modality used[
9]. Advanced signal processing approaches and machine learning algorithms improve the quality of the signal to decode mental states; this is the basis of a closed-loop BCI used to manage neurological disorders. Signal recording techniques include invasive and non-invasive techniques, with the latter comprising electroencephalography and functional near infrared spectroscopy, while the former comprises electrocorticography and intracortical recordings[
10]. Non-invasive techniques, such as EEG, have poor spatial resolution, while electrocorticography (EcoG) exhibits high spatial resolution because it records signals from the surface of the brain, and deep brain signals help in comprehending neural dynamics. The latest developments in the use of soft neurological interface materials indicate better biocompatibility, which enables stable recordings with minimum cellular damage. Memory networks can be modulated with the help of combined neural recording and invasive or non-invasive stimulations, which target specific neuronal networks that help in memory storage and retrieval[
11]. Advanced signal processing approaches, such as common spatial patterns and independent component analysis, help in improving the quality of the signal for better decoding of motor and other intentions.
In dementia, BCI interventions may be informed by the frequency-specific neural modulation approach. It has been noted that different frequency bands on the EEG may be related to different cognitive processes. For example, theta (4–7 Hz) and alpha (8–12 Hz) frequency bands may be closely associated with memory and attention, and the beta frequency band (13–30 Hz) may be relevant to executive function. Neurofeedback interventions may be based on the enhancement of alpha and theta frequency bands to improve attention and working memory, and the beta frequency band for the treatment of executive function. Moreover, gamma band activity, defined as greater than 30 Hz, has also been identified as a possible target for neuromodulation in Alzheimer’s disease, as indicated by early results showing the effectiveness of gamma entrainment in reducing amyloid load as well as enhancing network connectivity. Although the choice of frequency is heterogeneous across all studies, it has yet to be standardized, thereby requiring harmonization of the protocol as well as the development of individualized frequency approaches based on patient-specific neural profiles[
12]. Signal Processing includes processes such as digitization, transmission, preprocessing, feature extraction, pattern classification, and target command generation to operate external devices[
13]. It includes processes such as spike sorting, filtering, blind source separation (BSS), timeline-frequency analysis, and classification[
14].
Machine learning algorithms are used to classify brain signals and convert them into executable commands[
15]. Advances in deep learning and spatial filtering have progressively narrowed performance gaps between non-invasive and invasive BCIs, improving decoding accuracy as preserving safety[
12]. The output component consists of a robot’s arm to do the human’s intended action by the information already processed through the former interface. Feedback strategies enable users to adjust brain activity in real time, thereby creating a closed-loop system[
16]. The combination of artificial intelligence (AI) and machine-learning techniques such as convolutional neural networks, support vector machines, and transfer learning improves signal classification[
7].
Multivariate pattern classification approaches, including support vector machines (SVMs), have demonstrated capability of discriminating affective and cognitive brain states in real time. Such classifiers can generalize over distributed neural activation patterns and may allow detection of conditioned “yes” and “no” responses within affective BCI paradigms[
17]. Closed-loop BCI systems are particularly relevant to dementia research, as they enable adaptive cognitive practice and ongoing monitoring of neural activity[
18]. These systems may promote neural plasticity through reinforcing brain activity patterns linked to cognitive performance[
19]. In dementia research, most BCI systems employ non-invasive EEG paradigms due to safety considerations, whereas invasive interfaces remain largely experimental.
Neurobiological rationale for BCI use in dementia
Several neurobiological factors support using BCIs in dementia. Neuroplasticity is the basis of cognitive adaptation, and neurofeedback via BCI can potentially improve the activity of the remaining neural networks in early dementia. BCI-mediated feedback may reinforce residual neural networks and promote activity-dependent neuroplasticity in early disease stages[
20]. Large-scale network dysfunction in dementia particularly affects the default mode network (DMN), while compensatory recruitment of the frontoparietal control network (FPCN) and salience network (SN) may temporarily preserve executive functioning. Non-pharmacological cognitive interventions have been shown to improve compensatory inter-network connectivity, suggesting that neural engagement may strengthen compensatory network reorganization[
21]. The neurobiological mechanisms by which BCI-based interventions can target dementia are illustrated in Figure 3.
In the case of dementia patients, BCI interventions also focus on the use of targeted regions of the brain as well as the neural circuits which are affected or impaired as a result of the condition. Some of the regions which have been targeted in BCI interventions to treat patients with dementia include the hippocampus as well as the medial temporal structures. These regions play a critical role in the formation of memory and are the first to be affected as a result of Alzheimer’s disease. Other regions which have been targeted in the treatment of patients with dementia include the dorsolateral prefrontal cortex (DLPFC) as well as the parietal regions, which play a critical role in the working memory as well as the executive functions of the patients[
22]. Regarding the neural circuits which are affected as a result of the condition, the DMN, the FPCN, as well as the SN have been shown to be dysfunctional as a result of the condition. Moreover, new approaches involve investigating the modulation of limbic circuits and hippocampal-cortical connectivity to aid in memory consolidation and retrieval. This region-specific strategy will help to develop a mechanistic approach to targeted BCI interventions for dementia. Executive functions and working memory involve the coordination of FPCN, SN, and DMN. In cognitively demanding tasks such as n-back tasks, the deactivation of DMN is a prerequisite to continued task performance. At the same time, activation of dorsolateral prefrontal and parietal areas should occur. Inefficient deactivation of DMN and FPCN contributes to dementia pathogenesis[
22].
Targeted BCI treatments might help restore these networks by delivering real-time neurofeedback that reinforces desirable neural activity patterns and promotes adaptive reorganization of dysfunctional circuits. Classical conditioning, which underpins conditioning-based BCI paradigms discussed earlier, represents a form of implicit learning that depends on neural circuits often relatively spared in early and moderate Alzheimer’s disease. Evidence suggests that although acquisition may be slower, patients can still develop conditioned responses with sufficient exposure, supporting the theoretical feasibility of conditioning-based BCI paradigms in dementia.[
17] Furthermore, affective BCI paradigms that utilize emotion-related neural responses may lessen dependence on higher-order executive processing, thereby delivering potential usefulness in advanced dementia stages characterized by impaired task engagement[
4]. Importantly, emotional processing—particularly in the auditory domain—may remain relatively preserved in Alzheimer’s disease, and implicit semantic processing can persist even when conscious memory is impaired. Leveraging preserved affective and implicit pathways may therefore enhance the feasibility of emotion-based BCI communication systems in advanced dementia[
17]. BCIs can also be used as a digital biomarker tool, which can detect small changes in brain activity patterns before the onset of symptoms[
23]. BCIs additionally allow real-time monitoring of neural activity and may detect early neurophysiological changes that precede overt cognitive decline, offering more objective and continuous assessment than conventional diagnostic approaches.
Clinical applications
Early diagnosis and screening
Researchers have studied event-related potentials, particularly P300 responses and EEG biomarkers, as potential indicators of early cognitive impairment in Alzheimer’s disease and related dementias[
24]. These brain signals show the way attention and working memory are working, and they could be used as clear signs of early cognitive problems[
25]. Preliminary studies indicate possible feasibility; however, validation in large cohorts is lacking. Character-input and passive EEG-based BCIs have demonstrated moderate accuracy in preliminary studies. Emerging portable EEG-based platforms have been developed to facilitate rapid cognitive screening, including handheld two-channel EEG systems capable of mobile recording and machine-learning–based analysis within minutes. Fast periodic visual stimulation paradigms and virtual reality–integrated EEG systems have also been explored for early-stage dementia detection, demonstrating the feasibility of scalable, non-invasive screening tools in applied environments[
2].
Cognitive rehabilitation and neurofeedback
BCI-based neurofeedback enables individuals to modulate neural activity using live feedback. Among the elderly population, researchers introduced a new game incorporating a memory training element[
26]. The BCI training was shown to improve both attention and visuospatial and memory abilities with satisfactory usefulness and acceptability. In a cohort study of healthy, mostly Chinese-speaking elderly individuals, the same was repeated to explore the universal properties of the system and the training program across populations speaking different languages[
27]. They confirmed the potential of BCI training to improve cognition in both English- and Chinese-speaking elderly, showing its usability and acceptability across multiple populations. Short-term BCI-based cognitive training programs also demonstrated greater cognitive improvements than conventional interventions such as music or physical exercise in patients with dementia[
28]. Gomez-Pilar and colleagues developed a BCI system for neurofeedback (NF) training in healthy elders. There were five tasks, each with increasing difficulty. In these tasks, participants practiced logic and memory by studying, learning, and training. The feedback was displayed by an object moving on the screen during motor imagery tasks. Multisensory feedback during BCI-based cognitive training may further enhance memory storage by delivering semantically congruent stimuli across sensory modalities. The test and EEG results show that, after five practice sessions, participants’ cognitive abilities have improved, particularly in visual, language, and memory. Thus, EEG-based neurofeedback studies suggest potential improvements in attention, working memory, and executive functioning in individuals with mild cognitive impairment and early Alzheimer’s disease, although findings remain preliminary and heterogeneous[
24,
29].
Evidence from neuroimaging studies of cognitive interventions in dementia demonstrates increased functional connectivity within compensatory neural networks following short-term training, accompanied by modest improvements in working memory performance. These results indicate that network-level reorganization may underlie observed cognitive gains and provide a mechanistic framework for BCI-guided neurofeedback strategies[
22]. Although the observed cognitive benefits were modest and limited primarily to attention and working memory, with no reliable evidence of long-term improvement, the intervention is constrained by a lengthy training-and-testing procedure. This substantial time requirement limits its feasibility for immediate clinical application in patient populations.
Assistive technologies and functional support
BCI technology can assist patients with advanced dementia by enabling assistive communication and environmental control through interaction with external devices using brain signals[
24]. The traditional BCI approach uses instrumental learning and requires individuals to self-regulate brain activity. This might not be very effective for patients with Alzheimer’s disease because of the executive function and attention problems associated with the illness. A paradigm shift from traditional instrumental learning-based BCIs, which require active self-regulation of brain activity, to conditioning-based paradigms has been suggested, as introduced earlier. In conditioning-based BCIs, users generate neural responses through associative learning rather than deliberate control, which may be more suitable for patients with impaired executive function[
4,
17]. BCI systems allow real-time neural monitoring and translation of decoded signals into assistive device control, incorporating multimodal sensory feedback to form a functional closed loop that supports patient involvement with external environments[
12].
Disease monitoring
The employment of neural monitoring using wearable BCI systems might also be helpful in the longitudinal measurement of the progression of Alzheimer’s disease and the effectiveness of treatment[
30]. Real-time alert systems might also be developed to help caregivers manage patients suffering from dementia[
7]. Longitudinal EEG-based monitoring using AI systems might be helpful in the formulation of personalized models of disease progression. These models might enable clinicians to predict the progression of the disease[
31]. Patient-specific differences in the acquisition and extinction of conditioned responses may be related to disease severity. These differences possibly indicate the possibility that the performance of the condition using a BCI might be used as a biomarker for the progression of early Alzheimer’s disease[
17]. In comparison to other cognitive tests such as the Mini-Mental State Examination (MMSE) and Montreal Cognitive Assessment (MoCA) for evaluating mental functions, EEG-based BCIs may offer a complementary, and potentially more direct, measure of neurophysiological function compared to standard cognitive tests.
Neuromodulation
New technologies include the use of closed-loop stimulation and the application of stimulation in conjunction with neuromodulation therapies like transcranial and deep brain stimulation[
32]. A closed-loop BCI incorporates neural monitoring and stimulation techniques, which can actively adjust stimulation based on the brain’s activity. This aims to potentially enhance the efficacy of the therapy, although clinical validation remains limited[
33]. Non-invasive methods of closed-loop cortical stimulation have the capability of inducing neural reorganization, supporting their potential role in correcting dysfunctional neural circuits and augmenting cognitive function[
12].
Evidence from clinical studies
While recent BCI advances have primarily demonstrated clinical success in stroke, paralysis, and amyotrophic lateral sclerosis, applications in dementia remain largely exploratory, underscoring the early translational stage of cognitive BCI interventions[
12]. The existing clinical trials are predominantly small pilot studies (generally
n < 30) with short-term follow-up intervals[
34], limiting statistical power and generalizability. Major representative studies are listed in Table 1. Importantly, several methodological limitations substantially undermine confidence in the reported findings. Firstly, many of these studies lack adequate active control groups, as many BCI interventions are compared to passive or no intervention at all, making it difficult to disentangle actual neurophysiological effects from placebo or engagement-related effects[
35]. Secondly, blinding is rarely used, which creates the potential for performance and observer bias, especially for cognitively mediated outcomes. Thirdly, there is a large variability in cognitive outcome measures, with some studies using non-standardized neuropsychological tests. This creates a problem of cross-study comparison. Furthermore, cognitive testing is typically repeated, which may create a practice effect, especially in a study of short duration[
22]. In addition, the duration, intensity, and BCI paradigms used in the intervention protocols also differ, and all these factors add to the complexity in comparing and synthesizing the findings[
36,
37]. The absence of long-term follow-up makes it difficult to assess the long-term clinical benefits and disease-modifying properties. Therefore, the existing evidence base mainly supports the feasibility, and future studies need to be conducted as well-designed randomized trials to assess the true therapeutic value of BCI interventions for dementia patients[
11,
24].
Technical and clinical challenges
Technical challenges consist of signal noise, variability in neural responses, and calibration requirements. Additional engineering barriers comprise refining signal acquisition precision, optimizing decoding algorithms for more accurate brain-state classification, and developing more user-centric output devices suitable for cognitively impaired populations[
4]. Clinical challenges comprise a lack of uniform protocols and limited long-term evidence[
38]. Ethical issues include data privacy, patient autonomy, and availability[
39]. Additional barriers comprise prolonged calibration sessions due to interindividual neural variability, computing requirements of AI-based decoding, and low signal-to-noise levels in EEG recordings, all of which limit scalability within cognitively impaired populations[
7]. Practical limitations also include prolonged user training, low information transfer rates in imagery-based paradigms, and substantial calibration requirements, which together restrict usability in elderly and cognitively impaired populations. Although high-resolution modalities such as Functional Magnetic Resonance Imaging (fMRI) are valuable for identifying discriminative neural regions, their limited portability and high cost restrict clinical scalability, strengthening the need for translation into portable EEG- or Near-Infrared Spectroscopy (NIRS)-based systems for everyday dementia applications[
17].
Neurosurgical perspectives
From a neurosurgical perspective, BCIs represent a significant evolutionary step forward in the field of functional neurosurgery, especially in relation to the potential of BCI technologies to modulate cognitive networks at a circuit level[
39]. The development of implantable/hybrid BCI technologies offers the potential to both record and stimulate brain activity, including the potential to target memory-related networks, including hippocampal/limbic networks[
40]. The development of minimally invasive BCI technologies, such as endovascular BCI technologies and functional ultrasound-based BCI technologies, is focused on optimizing the tradeoff between the quality and safety of the procedure. The use of invasive BCI technologies for dementia patients has been associated with several challenges, such as patient selection and the risks involved. Dementia patients include elderly patients, and they are prone to having comorbidities. Thus, it is important to consider the disease state, the patient’s functional status, and the potential benefits of the device. Device-related issues, including long-term biocompatibility, signal stability, infection risk, and device longevity, are important, especially in light of the progressive nature of many of these diseases. The longevity of the device, as well as the possibility of revision, must be weighed against the potential benefits of the device. Ethical considerations are another concern in the neurosurgical community. Issues of informed consent are difficult in patients who are cognitively impaired, especially in patients who have fluctuating levels of decisional capacity. This necessitates structured consent processes, involvement of surrogate decision-makers, and ongoing reassessment of patient assent in longitudinal BCI use. Consequently, while neurosurgical BCI approaches hold promise for targeted neuromodulation in dementia, they remain investigational and require rigorous evaluation through ethically grounded and methodologically robust clinical trials.
Ethical aspects
The application of BCIs in the treatment of patients suffering from dementia also poses a number of ethical considerations beyond the general issue of privacy of information. Some of the considerations include the issue of informed consent. In the case of patients suffering from dementia, it is evident that the capacity of the patients may be fluctuating. This implies that informed consent poses a major challenge in the application of BCIs. In addition, the issue of ownership of neural data privacy also comes into play. BCIs have the ability to obtain neural data that is considered sensitive. In the future, the obtained neural data may be used to predict the cognitive status of the patient. In this respect, the application of neural data poses a number of ethical considerations, especially regarding the issue of data privacy, as well as the potential for misuse of the data obtained from the BCIs. This includes the issue of discrimination in insurance and employment[
7]. In addition, the possibility of cognitive and behavioral neuromodulation via closed-loop neuromodulation raises ethical issues about autonomy and identity. The interventions may inadvertently affect the personality, emotional state, and cognitive ability to make decisions, thus raising the issue of autonomy. The issue of equity and access to BCI technologies is also crucial, especially because the technologies may be costly and require extensive resources, thus further worsening the health inequalities gap[
11]. Therefore, ensuring equitable access to the technologies is critical to prevent the further health inequalities gap in dementia patients. The issue of cybersecurity breaches in BCI technologies, especially in the case of connected devices, is also critical, and unauthorized access to the data and control of the device may have serious implications[
41].
Future directions
Future research should focus on overcoming the methodological and translational limitations identified with the current BCI research. A priority should be given to the development of standardized cognitive outcome measures and training protocols to allow comparability between studies. Consensus should be established on the validation of neuropsychological outcomes to allow efficacy to be interpreted. Subsequently, well-designed multicenter randomized clinical trials with active comparators and blinding should be conducted to fully elucidate the therapeutic potential of BCI interventions in dementia patients. Long study designs should be considered to allow the long-term effects of BCI interventions to be evaluated as potential disease-modifying treatments. The development of technology should be directed toward the improvement of signal reliability, reduction of calibration time, and usability in the elderly and cognitively impaired. In addition, the integration of AI should be used for the improvement of decoding precision, as well as the development of adaptive patient-specific interventions, rather than the development of fully autonomous systems. At the same time, future work should be directed toward the development of multimodal BCI systems that combine BCI with neuroimaging and neuromodulation techniques. In addition, there is a need for the development of translational work regarding the cost-effectiveness of the technology. Finally, there is a need for the development of a framework regarding ethics, especially in the context of informed consent. These steps will be important for the translation of BCI technology in the context of dementia from proof of concept to clinically relevant application[
33].
Conclusion
Brain-computer interface technology represents an emerging approach for the diagnosis, monitoring, and treatment of dementia. Apart from direct cognitive enhancement, BCIs may serve as integrated therapeutic tools that improve autonomy, communication, and overall quality of life in patients with neurodegenerative disorders. The current evidence available indicates that non-invasive EEG-based BCI systems may offer a modest benefit in attention and working memory and appear to have potential as digital indicators for early cognitive impairment. However, most of the available evidence is from small pilot studies that have been acknowledged to have methodological heterogeneity and limited longitudinal follow-up, which precludes definitive conclusions about clinical efficacy. Future studies will involve large-scale well-designed randomized controlled trials with standardized cognitive tests and outcome measures. In addition, advances in AI assisted signal decoding and closed-loop neuromodulation may provide further opportunities to enhance the specificity of interventions. From a neurosurgical point of view, further development of implantable and hybrid BCI platforms may provide opportunities to modulate circuits. However, attention to issues of ethics, such as informed consent and access, will also be required to be addressed simultaneously with technology development. It should be emphasized that to overcome current methodological and ethical challenges, further research will be required to translate current preliminary feasibility findings to more clinically relevant interventions. Currently, BCI-based interventions for dementia can be considered investigational with a great deal of research to be done.
PDF
(1781KB)
