Motor imagery neurorehabilitation is a key technique in brain-computer interfaces (BCIs) for stroke survivors. It most often helps by training their brains to control robotic exoskeletons. But there’s a catch: staying focused during these mental workouts is exhausting. And when fatigue sets in, performance plummets.
We’ll talk about research that suggests a surprising solution—your heartbeat. Scientists have discovered that dips in heart rate variability (HRV) can predict when a user’s BCI control is about to slip. HRV is a good measure of nervous system flexibility.
By monitoring HRV in real time, future rehab devices could adapt before frustration kicks in, keeping therapy sessions effective and motivating.
Physiological measures in the closed-loop task
Seventeen participants did a MI task at different difficulty levels. Their task was to imagine either grasping with their left or their right hand. This is a common type of task amongst stroke recovery patients.
For the closed-loop task, researchers measured brain activity using a mobile EEG device. Based on the real-time EEG signal, participants got feedback on whether they succeeded in the task.
The same device was used to measure heart activity. Further down we’ll talk about how the heart rate was then extracted from the raw signal.
To move the hand in this task, we need to look at the sensorimotor rhythms in EEG (SMRs). These rhythms are expected in the frequency band of 8 to 12 Hz. We are expecting event-related desynchronization (ERD) of SMRs during this motor imagery task.
How do authors of the study measure SMR-ERD? – They look at the left side of the brain when right hand is imagined, and vice-versa for the other side. They picked electrodes C3 and C4, as well as their nearest neighboring electrodes.
After applying spatial filtering (Laplacian) to these electrodes, they calculated individual peak frequencies. And then used the Pfurtscheller and Aranibar method for final SMR-ERD.
Extracting HRV
From the filtered ECG signal, researchers extracted inter-beat intervals (IBIs). They also had to correct any abnormal beats.
The cleaned IBIs were resampled, bandpass-filtered to isolate high-frequency HRV (vagal tone). The standard deviation was calculated as the final measure.
Overview of the study design
Motor imagery study design.Before experiment day, participants had to come in and familiarize with the task. Then, on the second day researchers deployed the BCI system.
Day 2: We have two runs of 8.5 minutes. Both runs consist of multiple blocks, during which participants are instructed to relax or do the motor imagery task.
The two runs are different in task difficulty. Task difficulty is modulated in two ways. First one is the ration between relax and MI task blocks. Second way to modulate task difficulty is by decreasing SMR-ERD detection threshold.
Offline analysis of the collected data during motor imagery neurorehabilitation task
To understand whether BCI performance and HRV changed over time, and whether HRV could predict performance drops, the researchers used two approaches:
- Multilevel Modeling (tracking changes over time)
They treated each participant’s BCI performance and HRV as a series of connected data points. Since people perform differently, they used multilevel modeling, which accounts for individual differences while still spotting overall trends.
They checked if time and task difficulty affected performance and HRV. While making sure their results weren’t skewed by random fluctuations.
- Granger Causality (did HRV predict performance drops?)
They tested if changes in HRV happened before declines in BCI performance. Because brain and heart data can be non-stationary, they used a special method (Toda-Yamamoto test) to avoid false alarms.
BCI performance declined with task difficulty
BCI performance changes over time, in normal and hard task difficulty. Showing averaged values over participants.In the normal difficulty run, performance seemed stable. However, in the high difficulty part, performance decline is significant (−0.6%/10 s, total drop of ~24%). The decline simulates cognitive fatigue seen in stroke patients.
HRV as a predictive biomarker for BCI performance?
Now let’s look at the HRV trend:
HF-HRV changes. HF – high frequency component of HRV is often related to vagus nerve activity.HRV remained stable during the normal run but decreased in the high difficulty run. Which seems to follow the BCI performance metric.
These results align with the Yerkes-Dodson law. It states that excessive cognitive demand reduces parasympathetic (vagal) activity, leading to performance deterioration.
However, only the right side of the Yerkes-Dodson curve (high cognitive load) was tested. The left side (low arousal) needs investigation too.
Another future step for the authors of the study is doing the same analysis on stroke patients. Generalizability to stroke patients is unclear due to potential autonomic dysfunction post-stroke.
HRV can be a promising, easy-to-measure biomarker for BCI performance drops due to cognitive fatigue. This could lead to personalized neurorehabilitation protocols. The protocols could also adapt in real time, improving patient motivation and recovery.
Read the original paper linked here.
Curious if this setup is right for your study?
Our team works with researchers across neuroscience, sports science and clinical applications - we'll help you figure out if mobile EEG fits your study design.
