Computer Interaction Methods and Modes for Epilepsy Patients

Binlu Yang

College of Computer Science and Technology, Hainan University, Haikou, China

Keywords: Epilepsy, Human-Computer Interaction, Brain-Computer Interface, Artificial Intelligence, Smart Wearable

Devices.

Abstract: Epilepsy is one of the most common chronic neurological disorders worldwide. Traditional drug therapies

have limited effectiveness for such patients and long-term use may lead to side effects such as cognitive

impairment and metabolic disorders. In addition, patients face unpredictable seizure risks, traditional nursing

methods have limitations in evaluation and treatment effect, there are many technical aspects of epilepsy

patients that can be optimized, and innovative technological breakthroughs are urgently needed to break

through the existing bottlenecks. This paper reviews the application of brain-computer interfaces (BCIs),

multimodal interaction technologies, artificial intelligence (AI), eye-tracking, and smart wearable devices in

epilepsy management. It also proposes future research directions, including multimodal data integration,

nanoscale brain-computer interface (BCI) development, patient-participatory design, and ethical privacy

protection. These innovations aim to enhance diagnostic accuracy, enable personalized treatment, and

improve daily monitoring for epilepsy patients, thereby boosting their quality of life and advancing the

medical field toward greater intelligence and precision.

1 INTRODUCTION

Epilepsy is a prevalent neurological disorder

affecting approximately 70 million people globally

(Zhao et al., 2020). It is a chronic condition with a

broad impact, recognized by the World Health

Organization as one of the five key neurological and

psychiatric diseases requiring global prevention and

control (Beghi et al., 2019). Patients face challenges

such as seizure-related quality of life issues, reliance

on technological devices, and device design

inadaptability. Advancements in smart devices and

technologies aim to improve patients' lives,

prompting extensive research into epilepsy-specific

human-computer interaction (HCI), particularly in

interface and modality design. BCIs offer closed-loop

neural regulation by decoding brain signals, such as

implantable systems that suppress abnormal

discharges and reduce seizure frequency, and non-

invasive devices like lightweight (1.7g) head-

mounted microscopes that monitor neural activity and

blood oxygen metabolism, capturing pre-ictal

neurovascular signals. The integration of BCIs,

multimodal interaction technologies, and AI provides

new avenues for epilepsy monitoring and

intervention. Multimodal technologies enhance

system adaptability and patient compliance by

combining visual, auditory, and tactile data. For

instance, eye-tracking combined with ear-worn

electroencephalography(EEG) devices enables home

seizure warnings with 99.8% accuracy. AI algorithms

play a key role in multimodal data analysis, predicting

drug responses and optimizing doses through EEG

and genomic data fusion. Additionally, optogenetics

and gene therapy (e.g., adeno-associated virus-

delivered neuropeptide Y) offer molecular-level

seizure control. Despite progress, challenges remain

in multimodal data integration, device

miniaturization, biocompatibility, and ethical and

privacy risks. Future research should focus on

interdisciplinary innovation, such as nano-flexible

electrode development for reduced tissue damage and

long-term BCI implantation, patient-involved design

for improved device comfort and interfaces, and

ethical frameworks for neural data usage regulation.

This paper analyzes existing work, its effects, and

shortcomings, explores the characteristics and needs

of epilepsy patients, and discusses emerging research

directions.

Yang, B.

Computer Interaction Methods and Modes for Epilepsy Patients.

DOI: 10.5220/0014307100004718

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 2nd International Conference on Engineering Management, Information Technology and Intelligence (EMITI 2025), pages 28-36

ISBN: 978-989-758-792-4

2 EXISTING WORK AND ITS

EFFECTS

With the rapid development of artificial intelligence

and machine learning, the collaboration between

computer and medical fields have become

increasingly prominent in epilepsy research (Li et al.,

2024), as shown in figure 1.

Figure 1: Computer collaboration process (Li et al., 2024).

2.1 Brain-Computer Interface (BCI)

BCI devices collect brain activity data, transmitting it

to computer terminals for decoding algorithm

research. BCIs are highly valuable for real-time

epilepsy monitoring, prediction, and intervention,

offering personalized treatment (Chen, 2025).

Clinical techniques include EEG, magnetic resonance

imaging(MRI), and PET-CT (Zhang, 2022).

Implantable or non-invasive BCI devices (e.g.,

cortical electrodes, EEG headsets) continuously

collect EEG data, using AI algorithms to identify pre-

ictal abnormal discharges. Some BCIs can provide

warnings minutes to hours before seizures, helping

patients take safety measures.

2.2 Multimodal Interaction Technology

This technology combines EEG, EMG

(electromyogram), visual, and auditory data sources

to comprehensively reflect patient behavior,

improving epilepsy monitoring accuracy and severity

assessment. It also aids diagnosis based on seizure

observation and etiology (NICE, 2022). Adaptive

PageRank algorithms assess brain region importance,

considering interactions between regions, while

multi-kernel strategies address data heterogeneity by

integrating connectivity and node information for

classification (Frontiers, 2021). The system

comprises three main modules: EEG data acquisition,

cloud-based analysis, and monitoring/decision

support. EEG data is collected, transmitted wirelessly

to the cloud for analysis using machine learning, and

results are fed back to support treatment planning and

home epilepsy management, as shown in Fig. 2.

Figure 2: Basic components of the automatic seizure

detection system (Picture credit: Original).

Epileptic seizure physiological signals exhibit

spatiotemporal heterogeneity, making single-modal

data (e.g., EEG) susceptible to noise and insufficient

for comprehensive condition reflection. Multimodal

technology integrates EEG, EMG, motion sensors

(accelerometers, gyroscopes), visual behavior

analysis (cameras), and other physiological indicators

to build multidimensional feature models. For

example, combining EEG and motion analysis

enhances detection sensitivity for tonic-clonic

seizures to 100% through deep learning algorithms

like convolutional neural networks(CNN) and long

short-term memory networks(LSTM). Synchronized

monitoring of multiple physiological parameters,

such as skin conductance activity (EDA),

electrocardiogram (ECG), and EEG, captures

sympathetic nervous system excitement during

seizures, reducing missed detections (Ein Shoka et

al., 2023). Cross-modal alignment techniques, such as

ResizeNet networks, address cross-species EEG

signal differences and feature distribution shifts,

improving model generalization.

2.3 Artificial Intelligence (AI) in

Epilepsy Human-Computer

Interaction

AI algorithms play a crucial role in epilepsy

prediction, diagnosis, and rehabilitation. By

integrating EEG, ECG, and genomic data, they

construct personalized treatment models. Deep

learning models (e.g., CNN, LSTM) excel in EEG

signal analysis, identifying pre-ictal sharp waves with

98.72% sensitivity and 91.17% F1 score (Yu, 2021).

Transfer learning frameworks address inter-

Computer Interaction Methods and Modes for Epilepsy Patients

individual EEG differences, enhancing classification

performance. AI predicts drug responses and

optimizes doses, such as HCN1 channel-based

precision drug design. Google Health's AI model can

predict seizures an hour in advance with 85%

accuracy. AI also provides personalized treatment

plans by analyzing EHRs (electronic health records)

and multimodal data. Lin's team proposed a machine

learning-based prediction model for epilepsy patients

with cognitive impairment (Lin et al., 2021).

The suggested framework consists of three

collaborative stages:

Stage 1: IoT-based wearable medical sensors and

smartphones collect real-time data, connecting to

patients' EEG data collectors.

Stage 2: Cloud computing provides processing and

storage resources, receiving patient data via the

internet for classification and analysis. Abnormalities

are classified based on patient status, and results are

reported to healthcare providers, enabling early drug

intervention and real-time data updates.

Stage 3: Medical staff monitor patient records and

EEG data via cloud-based networks, review reports,

and take appropriate actions. BCI devices collect

brain activity data, transmitting it to computer

terminals for decoding algorithm research.

2.4 Eye-Tracking

Eye-tracking technology controls devices through eye

movement, offering precise input for epilepsy

patients unable to use hands or voice. Eye-trackers are

categorized into head-mounted and desktop types

(Zhang, 2022), as shown in Fig. 3.

Figure 3: Learning process of eye tracking (Zhang, 2022).

2.5 Smart Wearable Devices

Smart bands, watches, and other wearables record

physiological data like skin conductance,

temperature, pulse, and motion, using LSTM

networks to predict seizures. They provide real-time

monitoring and feedback for self-management. A

facial expression recognition system based on novel

flexible piezoresistive materials captures pressure

signals from facial muscle and skin deformation via

sensors in eyeglass legs, enabling emotion

classification (Zhang & Xing, 2025), as shown in Fig.

Figure 4: The workflow of our system (Zhang & Xing,

2025).

3 SHORTCOMINGS OF

EXISTING TECHNOLOGIES

3.1 Challenges of BCI Systems in

Epilepsy Applications

The spatiotemporal complexity of epileptic

discharges demands high algorithm accuracy for

signal decoding. The imbalance between inter-ictal

and ictal EEG data leads to high accuracy for non-

seizure data but poor recognition of seizure data,

causing false positives and missed detections. Future

algorithms must improve minority sample

recognition (Wang & Zhang, 2022). This imbalance

makes classifiers prone to false positives and missed

detections in practical applications, affecting the

accuracy and reliability of epilepsy seizure

prediction. In the future, more efficient and robust

algorithms need to be designed to address this issue,

enhancing the ability to identify minority class

samples, thereby improving the precision and

reliability of epilepsy seizure prediction (Xiaohui et

al., 2024).

Long-term device stability is also an issue, with

implantable electrodes suffering signal attenuation

EMITI 2025 - International Conference on Engineering Management, Information Technology and Intelligence

due to tissue encapsulation and non-invasive devices

being prone to motion artifacts. Moreover, individual

differences in seizure foci and types require highly

customized algorithms and stimulation parameters.

Pediatric patients' dynamically developing brains

further complicate BCI parameter adjustment,

necessitating personalized algorithms and timely

parameter updates.

3.2 Challenges in Multimodal Data

Fusion

Multimodal data heterogeneity is a significant

challenge. Differences in time resolution, spatial

features, and semantic associations across modalities

(e.g., EEG, MRI, ECG) complicate data fusion.

Ambiguous semantic links between genomic and

imaging data require complex algorithms, yet current

methods struggle to capture high-order heterogeneity,

risking information loss and poor fusion outcomes. In

clinical settings, modality missingness adaptation is

insufficient. Many multimodal systems rely on

complete data inputs, so missing key modalities due

to privacy, technical, or cost constraints degrades

system performance. This limits clinical application

and adversely affects diagnosis and treatment. Cross-

modal relationship modeling is further complicated

by difficulties in spatiotemporal alignment of

heterogeneous data. Existing algorithms often fail to

capture high-order associations between such data.

For example, structural and functional brain networks

provide different perspectives on epilepsy-related

structural changes, but common fusion models only

integrate information at a single granularity, ignoring

the multi-granular nature of brain networks and

leading to suboptimal fusion effects (Qi et al., 2024).

To address these challenges, researchers are

exploring new methods. For instance, the alternating

single-modal adaptation (MLA) method optimizes

each modality's encoder while integrating cross-

modal information to reduce interference and

enhance fusion performance. For modality

missingness, robust multimodal learning methods

like MoRA are being developed, which insert specific

modules to identify missing modalities and improve

model robustness under extreme missingness

conditions (Xiaohui et al., 2024).

3.3 Limitations of AI Epilepsy Early

Warning Systems

Real-time response delays and data bias/fairness

issues are two major challenges for AI in epilepsy

prediction. AI systems often rely on historical data for

predictions, which limits their ability to respond

dynamically to sudden epileptic events. For example,

epilepsy seizures are often sudden and unpredictable,

yet AI systems require time to process new data and

update models, potentially missing optimal

intervention timing. To enhance real-time

responsiveness, researchers are investigating new

algorithms and architectures, such as deep learning-

based real-time prediction models and edge

computing technologies to accelerate data processing.

Regarding data bias and fairness, training data is

often regionally and demographically skewed, with

most data originating from Western patients. This

results in reduced accuracy and applicability for

specific groups like children and ethnic minorities,

potentially exacerbating healthcare disparities. To

mitigate this, researchers recommend using more

representative datasets and developing adaptive

algorithms. Cross-institutional and international data

sharing can also help create more comprehensive and

balanced training datasets.

3.4 Impact of Different Environments

and Application Scenarios on

Eye-Tracking Effectiveness

Eye-tracking technology faces multiple challenges in

practical applications, including poor environmental

adaptability, insufficient real-time responsiveness,

and interference signals. Under complex

environmental conditions, such as low light or high

reflection, eye-tracking technology often

underperforms. In such cases, single-modal eye-

tracking data struggle to maintain system stability and

accuracy, necessitating multimodal data fusion

technology to enhance overall performance. By

integrating data from different modalities, such as

environmental sensor data or other biometric data, the

shortcomings of single-modal data can be effectively

addressed, improving the adaptability and robustness

of eye-tracking systems in complex environments.

The insufficient real-time responsiveness also

restricts the application of eye-tracking technology in

scenarios demanding high response speeds. For

instance, delays in eye-tracking can degrade user

experience and even pose safety risks in autonomous

driving or real-time interaction systems.

Interference signals are another significant

challenge for eye-tracking technology. Physiological

phenomena like eye jitter and blinking can cause data

interruptions or misjudgments, affecting system

accuracy and reliability. To tackle this issue,

researchers are developing advanced signal

processing algorithms to identify and filter out

Computer Interaction Methods and Modes for Epilepsy Patients

interference signals. Additionally, improving the

design of eye-tracking devices, such as using more

precise sensors and optimized optical systems, can

help reduce the impact of interference signals.

3.5 Hierarchical Security Protection

and Attack Path Analysis of

Wearable Devices

Missed detection of focal and non-motor seizures,

edge computing bottlenecks, and digital security risks

are key challenges for wearable devices in epilepsy

monitoring.

Wearable devices (e.g., wristbands) are highly

sensitive to tonic-clonic seizures, achieving 100%

detection accuracy when combining EEG and ECG,

but have low recognition rates for complex partial or

absence seizures, with missed detection rates

exceeding 50%. Focal autonomic seizures or absence

seizures lack significant physiological or motor

features, making them hard to detect. Existing devices

are less effective for these seizure types.

Edge computing bottlenecks constrain device

performance. High-precision AI models (e.g.,

Transformers) consume significant power on

wearables, creating a trade-off between performance

and battery life. Researchers are exploring efficient

algorithms and hardware optimizations to reduce

energy consumption and enhance computational

efficiency.

Digital security risks are a concern. Wearable

devices require layered protection from hardware to

application levels. Threats penetrating network

boundaries can compromise hardware and system

layers, jeopardizing application service and data

security. This may lead to digital asset risks and

potential post-attack denial by attackers (Zhao et al.,

2024). Device manufacturers need to strengthen

security measures, such as advanced encryption,

regular security patches, and user education.

4 CHARACTERISTICS AND

NEEDS OF EPILEPSY

PATIENTS

4.1 Disease Characteristics

Epilepsy patients face multiple challenges: seizures

are sudden and unpredictable, with irregular

occurrence and duration, potentially causing transient

loss of consciousness, motor control, or sensory

impairments. This increases the risk and difficulty of

using electronic devices. For instance, patients may

lose control during seizures or be unable to recall

operations afterward. Epilepsy is chronic and

recurrent, requiring lifelong management. About 30%

of patients develop drug-resistant epilepsy,

necessitating surgical or neuromodulation options.

Epileptic symptoms are diverse, including

generalized convulsions and brief loss of

consciousness, with varying impacts on daily life.

4.2 Physical and Cognitive Effects

Epileptic seizures pose significant physical injury

risks, such as falls and suffocation, potentially leading

to fractures or traumatic brain injuries. Frequent

seizures or medication side effects often result in

memory decline and attention deficits, particularly in

children. world health organization (WHO) data

indicates that 40%-60% of epilepsy patients

experience anxiety or depression, facing substantial

psychological and social pressures. Stigma and

psychological burdens lead many to conceal their

condition, while school and workplace discrimination

create additional challenges.

4.3 Core Needs of Epilepsy Patients

The needs of epilepsy patients stem from disease

characteristics and social biases. A comprehensive

support system covering "prevention-treatment-

integration" is required.

Epilepsy management must be multidimensional.

Precision medical support is essential, starting with

early diagnosis and correct classification.

Individualized treatment plans should be developed,

with drug-resistant patients trying alternative

therapies like neurostimulation or the ketogenic diet.

Treatment plans should be dynamically adjusted. For

safety and emergency care, patient environments

should be safety-adapted, and scientific first aid

knowledge should be promoted. Special groups have

unique needs: children require cognitive

rehabilitation and personalized educational support

(e.g., Cambridge University's "EpiSchool" AI

platform for customized learning paths); women of

childbearing age need pregnancy medication

guidance and genetic risk counseling; elderly

patients, often with multiple comorbidities, require

hospital-based comorbidity management and family

monitoring with home safety modifications.

EMITI 2025 - International Conference on Engineering Management, Information Technology and Intelligence

5 DISCUSSION AND ANALYSIS

5.1 Multimodal Interaction in Home

and Medical Settings

5.1.1 Multimodal Data Integration

Real-time monitoring data from EEG, ECG, EMG,

and wearables (smart bands/clothing). Cameras

capture limb movements, and voice interactions are

recorded. Environmental data such as temperature,

humidity, and intensity of illumination in homes and

public places are also collected. Integrating these data

sources creates a comprehensive patient data profile

throughout the life cycle.

5.1.2 Subjective Feedback

Voice diaries and emotion recognition (via NLP to

detect anxiety/depression) from patients and families.

Combined with genetic data, medication history, and

multimodal monitoring results, dynamic medication

suggestions (e.g., dosage adjustment or drug

switching) are generated. Timely medication

reminders, sleep advice, and seizure trigger

avoidance (e.g., staying up late, strong light

stimulation) are provided to help patients establish

regular routines. Disease causes, first aid measures,

and treatment progress are explained in layman's

terms to alleviate patients' fears stemming from

misunderstandings.

5.1.3 Smart Home Integration

During a seizure, AI systems recognize falling

motions via cameras, automatically shutting off gas,

dimming lights, and activating emergency calls.

Smart home devices provide vibration or voice

prompts to patients ("You are about to erupt. Please

sit down ") and send location information to family

members.

5.2 Emerging Research Directions

5.2.1 Multisensor Fusion for Early Warning

Devices integrate motion sensors (detecting abnormal

convulsions), skin conductance sensors (monitoring

stress levels), and microphones (identifying abnormal

breathing sounds). Edge computing enables real-time

analysis and alarm triggering. Upon detecting

abnormal EEG signals, emergency procedures are

initiated, contacting emergency contacts or medical

institutions automatically. Seizure times and

symptoms are recorded for doctors' reference.

5.2.2 AR/VR + Dialogue Robots for

Rehabilitation Training

AR simulates high-risk scenarios (e.g., crossing

roads) to train patients in self-protection actions upon

recognizing premonitory symptoms. VR provides

relaxing environments to alleviate anxiety (e.g.,

meditation forests) and simulates social interactions

to boost patients' confidence and social engagement.

5.2.3 Doctor-Patient Remote Collaboration

Patients can film seizure videos with their

smartphones. AI automatically marks key frames

(e.g., tonic-clonic phases), enabling doctors to

diagnose quickly combining voice descriptions. This

is applicable for emergency handling and daily

monitoring, reducing safety risks for patients going

out alone and saving time for both doctors and

patients. It also provides more convenient and

equitable medical support for remote patients.

5.2.4 Nano-BCI

Nano-particles or flexible electronics enable non-

invasive, high-precision monitoring, reducing the

immunoreaction to implantable BCIs and increasing

patient acceptance.

5.2.5 Brain-Cloud Interface

Epilepsy patients can upload EEG data to the cloud in

real-time. After verification, the system synchronizes

data to doctors' platforms. Qualified physicians can

access and analyze global patient data online for

monitoring, diagnosis, and treatment.

5.3 Patient-Centered Design

Epilepsy can cause cognitive impairments and

tendencies toward depression and anxiety. It is crucial

not only to enhance technologies for predicting and

diagnosing epilepsy but also to monitor patients'

emotions in real-time to prevent worsening

conditions or impulsive negative behaviors.

Automatically adjusting interaction methods and

feedback based on patients' physiological and

psychological characteristics to provide customized

user experiences represents a future research

challenge.

Computer Interaction Methods and Modes for Epilepsy Patients

5.3.1 Emotion Recognition and Guidance

Natural Language Processing (NLP) analyzes

anxiety, depression, or loneliness in patients' speech,

offering immediate comfort such as guided

mindfulness exercises, relaxation techniques, or

referrals to professional psychological resources. It is

necessary to train models to distinguish between

pathological and psychological emotional changes to

avoid misjudgments. For patients with social

limitations due to their condition, robots can reduce

loneliness through daily conversations and encourage

emotional expression.

5.3.2 Personalized Adaptation

Customize dialogue content based on patients' age,

severity, and cultural background. Combine voice,

text, and visual interfaces (e.g., animated breathing

guidance) to accommodate different communication

preferences. Incorporate patient feedback during

development to optimize interface and functionality

adaptation.

5.3.3 Affective Computing and AI

Companionship

Voice emotion recognition (e.g., tone, speed) and

facial expression analysis enable AI chatbots to

provide real-time psychological support. Combined

with soothing music, dynamic lighting, and tactile

feedback (e.g., pressure blankets), anxiety levels can

be reduced.

5.4 Ethics and Privacy Protection

Given the global nature of epilepsy patients, data

encryption standards (e.g., EU AI Act) should be

established to build cross-cultural ethical consensus.

5.4.1 Privacy Protection

Encrypt storage of patient health data (e.g., seizure

records, medication information) and biometric data

(e.g., EEG) to comply with medical data regulations

(e.g., HIPAA, GDPR). Studies indicate that EEG

signals collected under identical stimuli can identify

individuals with near 100% accuracy. Leaked

biometric information from such stimuli can still

identify individuals (Ruiz-Blondet et al., 2016). At

the 2012 USENIX Security Symposium, Professor

Ivan Martinovic from Oxford University introduced

"brain spyware" that collects BCI data to steal

information such as addresses, birthdates, credit card

numbers, and acquaintances by analyzing users'

visual stimulation responses (Martinovic et al., 2012).

Hackers may alter BCI data to manipulate external

devices for illegal purposesErro! A origem da

referência não foi encontrada. (Chen, 2025).

5.4.2 Liability Boundaries

Clearly define robots as auxiliary tools, not

substitutes for professional medical advice.This

fundamental distinction is critical to ensuring patient

safety and maintaining the integrity of medical

practice. To underscore this, explicit and prominent

risk warnings must be incorporated into the

operational protocols and user interfaces of these

robotic systems. For instance, a warning such as “In

the event of any discomfort or unwell symptoms,

contact a doctor immediately” should be readily

accessible and visible to users at all times (Schermer,

2009). Such warnings aim to prevent users from over

- relying on the robotic systems and neglecting the

necessity of professional medical diagnosis and

intervention when necessary.

5.4.3 Cultural Sensitivity

Avoid misunderstandings due to cultural differences

(e.g., epilepsy stigmatization in some regions) by

designing inclusive dialogue logic. Cultural

differences can significantly influence how health

conditions are perceived, discussed, and addressed.

For example, in some regions, epilepsy may be

stigmatized due to traditional beliefs, myths, or lack

of awareness about its true nature as a neurological

disorder. Such stigmatization can affect patients'

willingness to seek help, adhere to treatment, and

discuss their condition openly. To address these

challenges, it is crucial to design inclusive dialogue

logic within healthcare technologies. This involves a

thorough understanding of diverse cultural

perspectives, values, and beliefs related to health and

illness. By incorporating this understanding into the

design of conversational interfaces, people can create

more empathetic, appropriate, and effective

interactions that respect cultural differences. By

prioritizing cultural sensitivity in the design of

healthcare technology dialogue logic, people can

reduce the risk of misunderstandings, enhance patient

trust, and improve the overall effectiveness of

healthcare interactions, ultimately contributing to

more equitable and accessible healthcare for diverse

populations.

EMITI 2025 - International Conference on Engineering Management, Information Technology and Intelligence

5.4.4 Fairness and Inclusiveness

Advocate for the inclusion of advanced devices (e.g.,

BCIs) in medical insurance to reduce the burden on

low-income families. Expand rural coverage through

telemedicine. Prevent predictive failures for specific

groups (e.g., children or ethnic minorities) due to

training data biases.

6 CONCLUSION

This paper systematically reviews the current state,

applications, and future directions of human-

computer interaction technologies for epilepsy

patients. It examines challenges faced by epilepsy

patients, including seizure unpredictability and

limitations of traditional care. It also reviews the

applications and limitations of BCIs, multimodal

interaction technologies, AI, eye-tracking, and smart

wearables in epilepsy monitoring, warning, and

intervention. Furthermore, it proposes future research

directions, including multimodal data integration,

nano-BCI development, patient-centered design, and

ethical and privacy protection. By integrating

technology and humanistic concern, it aims to

establish a comprehensive epilepsy management

ecosystem covering monitoring, intervention, and

feedback, providing full-cycle health management for

patients.

BCIs, multimodal interaction technologies, and

AI offer a transformative path from "passive control"

to "active intervention" in epilepsy treatment,

enhancing monitoring accuracy and intervention

timeliness. However, clinical application challenges

persist, including multimodal data fusion complexity,

device long-term stability, real-time response delays,

and ethical and privacy risks. Human-computer

interaction technologies in epilepsy prediction still

face challenges such as real-time response delays and

data bias/fairness issues. Overcoming these requires

technological innovations like more efficient

algorithms and architectures, as well as social and

policy support, including data sharing and fairness

standard development. With advancements in brain

science and AI, the future promises a safer, more

precise, and inclusive epilepsy management system,

achieving comprehensive support for "prevention-

treatment-integration."

REFERENCES

Beghi, E., Giussani, G., Nichols, E., et al.: 'Global, regional,

and national burden of epilepsy,1990 2016: a

systematic analysis for the Global Burden of Disease

Study 2016', The Lancet Neurology, 2019, 18(4): 357-

375

Chen, T. L.: 'Research on the Use and Protection of Brain-

Computer Interface Data', Journal of Xichang College

(Social Science Edition), 1-11, 2025

Ein Shoka, A.A., Dessouky, M.M., El-Sayed, A., et al.:

'EEG seizure detection: concepts, techniques,

challenges, and future trends', Multimed Tools Appl,

82, 42021-42051, 2023

Frontiers in Neuroscience.: ‘Epilepsy Detection and

Recognition Based on Multimodal Signals’. (IF 3.2)

Pub Date: 2021-09-29

Li, Q., Li, P.F., D., Zhao, Z., et al.: 'An Epileptic EEG

Imbalanced Classification Method Combining

Reinforcement Learning with Swarm Intelligence

Algorithms', Journal of Chongqing University of

Technology (Natural Science), 2024, 38(12): 110-123

Lin, F., Han, J., Xue, T., et al.: 'Predicting cognitive

impairment in outpatients with epilepsy using machine

learning techniques', Scientific Reports, 2021, 11(1):

20002

Martinovic, I., Davies, D., Frank, M., et al.: 'On the

feasibility of Side-Channel attacks with Brain-

Computer interfaces', 21st USENIX Security

Symposium (USENIX Security 12), 2012: 143-158

NICE guideline[NG217].Epilepsies in children, young

people and adults[EB/OL](27 April 2022)[9 May

2022].https://www.nice.org.uk/guidance/ng217.

Qi, X. Y., Ding, W. P., Ju, H. R.: 'Seizure recognition

method based on multimodal multi-grain fusion

network', Data Acquisition and Processing, 2024,

39(3): 710-723

Ruiz-Blondet, M.V., Jin, Z., Laszlo, S.: 'CEREBRE: A

novel method for very high accuracy event-related

potential biometric identification', IEEE Transactions

on Information Forensics and Security, 2016, 11(7):

1618-1629

Schermer, M.: 'The mind and the machine: on the

conceptual and moral implications of brain-machine

interaction', Nanoethics, 2009, 3(3): 217-230

Wang, Y., Zhang, L.: 'A novel clustering-based exploratory

SMOTE algorithm', Journal of Chongqing University

of Technology (Natural Science), 2022, 36(04): 187-

195

Xiaohui, Z., Jaehong, Y., Mohit, B.: 'Multimodal

Representation Learning by Alternating Unimodal

Adaptation' arXiv:2311.10707. 2024

Yu, Y. J.: ‘Intelligent Epilepsy Detection and Recognition

Based on Multimodal Signals’. Hangzhou Dianzi

University. 2021

Zhang, H.Y.: 'Research on Eye Movement Signal

Processing and Classification Algorithms for Cognitive

Tasks in Epileptic Patients', University of Chinese

Academy of Sciences (Xi'an Institute of Optics,

Computer Interaction Methods and Modes for Epilepsy Patients

Precision Mechanics and Physics, Chinese Academy of

Sciences), 2022

Zhang, J., Xing, X.: 'Smart Glasses Facial Action

Recognition System Using Improved Meta-Learning

Network', Microcomputer Information, 1-10, 2025

Zhao, G., Zheng, Y., Tao, Z. L.: 'System risk assessment

analysis for intelligent wearable devices', Information

Security for Computer Networks, 2024, 24(10): 1595-

1603

Zhao, J., Zhao, D., Shi, L., et al.: 'Multilayer weighted

integrated self-learning algorithm for automatic

diagnosis of epileptic electroencephalogram signals',

Computational Intelligence, 2020, 38(1): 3-19

EMITI 2025 - International Conference on Engineering Management, Information Technology and Intelligence