Drowsiness and Fatigue Detection in Rehabilitation with Passive BCI

Is your brain ready to learn right now?

Intro

Imagine a stroke survivor sitting in front of a screen, trying to mentally “move” a paralyzed hand. This kind of training, known as motor imagery-based Brain-Computer Interface (MI-BCI) rehabilitation, helps the brain relearn movement by repeatedly activating motor-related brain networks to facilitate restorative neuroplasticity.

But after 20–30 minutes of intense concentration, a common challenge arises: the task becomes repetitive, attention drops, and the patient becomes mentally tired. To capture this mental effort, patients wear a specialized cap lined with sensitive electrodes that perform Electroencephalography (EEG), a non-invasive method of monitoring the brain's internal electrical conversations. As cognitive fatigue sets in, these EEG signals begin to shift, specifically, research shows a noticeable decrease in "beta band" activity in the frontal and central brain regions, which is a direct neurological indicator of lowered brain arousal. When this happens, the patient struggles to complete the mental tasks, the BCI system's classification accuracy deteriorates, and the rehabilitation session becomes less effective.

What if the therapy system could notice this fatigue in real time?

The goal is simple but powerful: provide "context-awareness" that makes rehabilitation smarter. By dynamically adapting to the patient’s real-time mental state, adaptive BCIs can prevent compromised performance, manage frustration, and ultimately improve the outcomes of stroke recovery.

Most people imagine a brain-computer interface as a system that lets someone control a cursor, robotic arm, or wheelchair using brain activity. That kind of system is usually an active BCI: the user intentionally produces a mental command

A passive BCI is different.

A passive BCI does not wait for a deliberate command. Instead, it continuously monitors the user’s brain activity, usually through electroencephalography, or EEG, to estimate internal mental states.

Such mental states include:

Fatigue
Drowsiness / Vigilance
Attention
Mental workload
Frustration
Engagement

In rehabilitation, this is especially useful because the patient may be trying to perform an active task, such as motor imagery, while the passive BCI monitors whether the patient is still in the right mental state to benefit from training. Stroke survivors, in particular, are often unable to maintain the same mental state throughout a session, making it crucial to detect when cognitive fatigue sets in.

What Is a Passive BCI?

Why Fatigue Matters in Rehabilitation

Rehabilitation after stroke is not only a physical process. It also depends on attention, motivation, mood, cognitive effort, and the patient’s ability to stay engaged.

A tired patient may still be sitting in the session, but their brain may no longer be learning effectively. Fatigue can reduce attention, increase frustration, lower mood, and weaken the quality of interaction with the BCI system.

This matters because motor rehabilitation often depends on repeated practice. If the system keeps training while the patient is mentally exhausted, the session may become less efficient. In some cases, the patient may also become discouraged because the task feels harder and the feedback becomes less reliable.

Passive BCI offers a way to make therapy more responsive. Instead of using a fixed schedule, the system could adapt based on the patient’s current brain state.

Possible adaptations include:

Suggesting a short break
Reducing task difficulty
Recalibrating the classifier
Switching to a different training mode
Giving motivational feedback
Ending the session before performance collapses

In this way, fatigue detection is not just about measuring tiredness. It is about protecting the quality of rehabilitation.

EEG measures electrical activity from the scalp. These signals can be analyzed in frequency bands, which are often linked to different brain states. During fatigue or drowsiness, several EEG patterns may change.

Theta Activity:
4–8 Hz Theta activity is often associated with drowsiness, reduced alertness, and mental fatigue (AL-Quraishi et al., 2018). When fatigue increases, theta activity may rise. In a rehabilitation session, increasing theta may suggest that the patient is becoming sleepy or mentally overloaded.

Beta Activity:
13–30 Hz Beta activity is often associated with active engagement and arousal. During fatigue, beta activity may decline. A drop in beta activity may suggest that the patient is becoming less alert or less actively engaged in the task, as beta rhythms act as direct neurological indicators of brain arousal levels.

Alpha Activity:
8–13 Hz Alpha activity is more complicated. It is tempting to say that alpha always increases during fatigue, but the scientific picture is not that simple. Some studies report alpha increases, others report decreases, and some describe a shift in where alpha appears across the scalp. One important idea is alpha anteriorization. This means alpha activity may decrease in posterior brain regions, such as occipital and parietal areas, while increasing in frontal regions.

This complexity is important. A good passive BCI should not rely on one simplistic marker. Instead, it should combine multiple EEG features and interpret them carefully.

What Does a Tired Brain Look Like?

From EEG to Fatigue Detection

A passive BCI fatigue-detection system operates through a sequential data-processing pipeline to achieve context-awareness.

Step 1: Record EEG
The patient wears an EEG cap, typically configured according to the international 10–20 system, during a rehabilitation session. These sensors capture ongoing neuroelectrical brain activity while the patient performs an active therapeutic task, such as kinesthetic motor imagery (MI) of a paralyzed limb.

Step 2: Clean the Signal
EEG data is inherently noisy and highly susceptible to artifacts from eye blinks, muscle movements (EMG contamination), and external electrical noise. Before analysis, the signal must be rigorously cleaned using band-pass filters and advanced artifact removal techniques like Independent Component Analysis (ICA) to isolate and discard ocular and muscular artifacts. Additionally, spatial filtering methods like the surface Laplacian or canonical correlation analysis can be applied to specifically mitigate circumferential muscle contamination.

Step 3: Extract Features
The system then calculates robust signal features from the cleaned EEG. This typically involves using a Fast Fourier Transform (FFT) to convert the time-domain signal into the frequency domain, allowing the extraction of spectral power within specific frequency bands—such as theta (4–8 Hz), alpha (8–13 Hz), and beta (13–30 Hz). Features derived from frontal and central cortical electrodes have been shown to be especially predictive for estimating mental fatigue.

Step 4: Classify Mental State
A machine-learning model uses these extracted features to continuously estimate the user's underlying mental state on a single-trial basis. Classifiers such as shrinkage Linear Discriminant Analysis (LDA), Support Vector Machines (SVM), or Riemannian Minimum Distance to Mean (MDM) are frequently employed to robustly discriminate whether the patient is attentive, fatigued, drowsy, or frustrated.

Step 5: Adapt the Rehabilitation Session
If cognitive fatigue is detected, the adaptive BCI system can dynamically respond to prevent compromised classification performance. It may pause the session, extend inter-trial or inter-session rest intervals, lower the task difficulty, recalibrate the learning algorithm to the current mental state, or recommend ending the training for the day to protect the quality of rehabilitation.

The key idea is that the BCI becomes context-aware: it registers not only the voluntary motor command the patient is trying to execute, but also uses physiological markers to understand if the patient's brain arousal level is ready to optimally facilitate restorative neuroplasticity.

Passive BCI fatigue detection is promising, but it is not simple.

1. EEG Is Noisy
EEG signals are inherently noisy and highly susceptible to ocular artifacts (blinking), electromyographic (EMG) contamination from muscle movement, electrode shifts, and external environmental noise. Before meaningful states can be extracted, the signal must be rigorously cleaned using advanced spatial filtering and artifact removal techniques like Independent Component Analysis (ICA).

2. Fatigue Is Not One Thing
A tired patient may be experiencing sleepiness (reduced vigilance), acute task frustration, cognitive overload, post-stroke depression (PSD), or sheer physical weakness. These mental states can overlap significantly, yet they possess distinct cortical correlates. For instance, frustration relies heavily on posterior alpha and frontal beta band activity, whereas simple cognitive fatigue strongly impacts frontal and central electrodes across multiple bands.

3. People Differ
There is significant inter-participant variability in the cortical regions and frequency ranges that are most predictive of changes in mental state. The same EEG pattern may not mean exactly the same thing for every patient. Thus, a robust system requires deep personalization, often utilizing participant-dependent classifiers and individualized frequency band clustering rather than relying on generic physiological thresholds.

4. Alpha Is Complicated
Alpha activity (8–13 Hz) does not always change in a uniform direction during fatigue or cognitive exertion. While some studies report alpha increases during fatigue, others report decreases, and some describe a spatial shift known as alpha anteriorization, where alpha activity decreases in parieto-occipital areas while simultaneously increasing in frontal regions.

5. Clinical Use Requires Caution
A rehabilitation system should not stop or drastically change therapy based on weak or isolated physiological evidence, as specific EEG markers (like relative beta power drops) sometimes exhibit statistically significant but low overall correlations with performance drop-offs. To be effective and safe, passive BCIs must combine continuous EEG monitoring with behavioral data, real-time performance metrics, and a comprehensive battery of clinical assessments to inform clinical judgment.

Challenges
and Limitations

Call
to Action

Traditional active BCIs often discard valuable psychological data by focusing solely on voluntary commands. However, because cognitive fatigue lowers brain arousal and BCI accuracy, adapting therapy to the user's mental state is vital to maximize restorative neuroplasticity.

Open questions to explore:

Optimal Pauses: Identifying data-driven thresholds for breaks is critical, as severe fatigue typically sets in after just 20–30 minutes of training.
Distinguishing States: Systems must reliably separate acute task fatigue from frustration and chronic Post-Stroke Depression (PSD) to maintain patient engagement.
Adaptation Control: It remains open whether adaptive BCIs should automatically lower task difficulty and recalibrate algorithms, or if patients should explicitly control these changes.
Home Care & Privacy: While passive BCIs could safely guide remote sessions by preventing mental overload, continuous monitoring extracts intimate cognitive details that necessitate robust privacy frameworks.

The next generation of rehabilitation technology may not simply ask, “What movement is the patient imagining?” It may also ask, “Is the brain ready to learn right now?”

Suggested Experiments

To advance this field and answer the open questions above, several specific experimental paradigms should be explored:

Adaptive vs. Fixed-Schedule Trials: Conduct Randomized Controlled Trials (RCTs) comparing standard MI-BCI therapy (using fixed 60-minute training blocks) against an adaptive pBCI system that dynamically adjusts inter-trial rest periods or decreases task difficulty in real-time whenever significant beta-band power drops are detected.
Multi-Class Continuous State Estimation: Move beyond simplistic binary classifications (e.g., alert vs. fatigued). Future experiments should utilize regression-based machine learning approaches or multi-class models to track the natural, continuous fluctuations of fatigue, frustration, and attention simultaneously on a single-trial basis.
Combining pBCI with Brain Priming: Investigate sequential therapies where a pBCI system detects high cognitive fatigue and subsequently triggers the application of Non-Invasive Brain Stimulation (NIBS), such as transcranial direct current stimulation (tDCS), to selectively modulate cortical excitability, increase brain arousal, and prime the neural networks for continued training.

Future Applications

Hybrid BCIs: The development of integrated systems that combine active BCI (decoding the user's voluntary motor imagery) with passive BCI (monitoring their physiological and psychological disposition). This hybrid approach would ensure the system only triggers therapeutic mechanical movement (like a robotic exoskeleton) when the user is generating the correct motor command and is in an optimal state of engagement.
Neurophysiology-Guided Personalized Rehabilitation: Creating holistic, all-encompassing rehabilitation programs where a patient's daily therapy is customized based on a comprehensive admission profile of their specific motor, cognitive, and affective deficits. Using this profile, a pBCI would continuously adjust the dosage, intensity, and frequency of the ongoing training based on real-time neurophysiological biomarkers of fatigue and mood to prevent recovery plateaus.

References List

electrode cap on a person's head measuring their brain dynamics.jpg

1. Al-Quraishi, M. S., Elamvazuthi, I., Daud, S. A., Parasuraman, S., & Borboni, A. (2018). EEG-based control for upper and lower limb exoskeletons and prostheses: A systematic review. Sensors, 18(10), Article 3342.

2. Foong, R., Ang, K. K., Quek, C., Guan, C., Phua, K. S., Kuah, C. W. K., Deshmukh, V. A., Yam, L. H. L., Rajeswaran, D. K., Tang, N., Chew, E., & Chua, K. S. G. (2020). Assessment of the efficacy of EEG-based MI-BCI with visual feedback and EEG correlates of mental fatigue for upper-limb stroke rehabilitation. IEEE Transactions on Biomedical Engineering, 67(3), 786–795.

3. Frolov, A. A., Mokienko, O., Lyukmanov, R., Biryukova, E., Kotov, S., Turbina, L., Nadareyshvily, G., & Bushkova, Y. (2017). Post-stroke rehabilitation training with a motor-imagery-based brain-computer interface (BCI)-controlled hand exoskeleton: A randomized controlled multicenter trial. Frontiers in Neuroscience, 11, Article 400.

4. Hinss, M. F., Jahanpour, E. S., Somon, B., Pluchon, L., Dehais, F., & Roy, R. N. (2023). Open multi-session and multi-task EEG cognitive dataset for passive brain-computer interface applications. Scientific Data, 10(1), 1–14.

5. Mane, R., Chouhan, T., & Guan, C. (2020). BCI for stroke rehabilitation: Motor and beyond. Journal of Neural Engineering, 17(4), Article 041001.

brain with a period of 300 miliseconds reaction.jpg

6. MNE Developers. (2026). Tutorials — MNE 1.12.1 documentation. MNE-Python.

7. MNE-Python Contributors. (2026). mne-tools/mne-python: MNE: Magnetoencephalography (MEG) and Electroencephalography (EEG) in Python [Source code]. GitHub.

8. Myrden, A., & Chau, T. (2017). A passive EEG-BCI for single-trial detection of changes in mental state. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 25(4), 345–353.

hi, M. S., Elamvazuthi, I., Daud, S. A., Parasuraman, S., & Borboni, A. (2018). EEG-Based Control for Upper and Lower Limb Exoskeletons and Prostheses: A Systematic Review. Sensors, 18(10), 3342.