User Tutorial:Introduction to the Mu Rhythm
In human EEG, primary sensory or motor cortical areas typically exhibit rhythmic activity at a frequency of approximately 8-12 Hz when they are not processing sensory information or producing motor output. This activity is called mu rhythm, and is thought to be produced by interactions between the thalamus and the cortex. Computer-based analyses have demonstrated that mu rhythm activity consists of a variety of different 8-12 Hz rhythms that are distinguished from each other by precise location, precise frequency, and/or typical relationship to concurrent sensory input or motor output.
Several factors suggest that mu rhythm activity could be a good carrier for BCI-based communication. These rhythms are associated with those cortical areas that are most directly connected to the brain's normal motor output channels. Movement or preparation for movement is typically accompanied by a decrease in mu activity over sensorimotor cortex, particularly contralateral to the movement. This decrease has been labeled "event-related desynchronization" or ERD by Pfurtscheller (Pfurtscheller, G.: EEG event-related desynchronization (ERD) and event-related synchronization (ERS). In: E. Niedermeyer, F.H. Lopes da Silva (eds.) Electroencephalography: basic principles, clinical applications and related fields, 4th edition, pp. 958–967. Williams and Wilkins, Baltimore, MD (1999)). Its opposite, rhythm increase, or "event-related synchronization" (ERS) occurs in the post-movement period and with relaxation. Furthermore, and most relevant for BCI applications, ERD and ERS occur also with motor imagery (i.e., imagined movement); they do not require actual movement. Thus, they can occur independent of activity in the brain's normal output channels of peripheral nerves and muscles, and could serve as the basis for a BCI.
The figure displays examples of modulated mu rhythm signals (modified from ).
- A,B: Topographical distribution on the scalp of the difference (measured as (the proportion of the single-trial variance that is due to the task)), calculated for actual (A) and imagined (B) right-hand movements vs. rest for a 3 Hz band centered at 12 Hz.
- C: Example voltage spectra for a different subject and a location over left sensorimotor cortex (i.e., C3) for comparing rest (dashed line) and imagery (solid line).
- D: Corresponding spectrum for imagery vs. rest. Signal modulation is focused over sensorimotor cortex and in the alpha- and beta frequency bands associated with mu rhythm activity.
In its center, the figure displays a human brain viewed from above, with the frontal direction pointing downward. At the top of the figure, a vertical cross-section of the brain is depicted, taken along motor resp. sensory cortex. On the left side, the motor cortex is displayed in red, associated with a "motor homunculus" illustrating which regions are allocated for controlling the respective limb and facial muscles. Similarly, on the right, sensory areas are illustrated by a "sensory homunculus" indicating which regions are allocated for processing sensory information from the respective parts of the body. The separation between motor and sensory cortices is a major landmark of brain anatomy, and is called the central sulcus or rolandic fissure.
The mu rhythm originates from the hand area of the motor cortex. Also, there is a similar rhythm originating from the motor cortex' foot area, which is located between hemispheres.
The mu rhythm's geometric source character is that of a dipole, with the dipole moment pointing perpendicular to the folded cortical surface. Thus, the orientation of the dipole moment is determined by the dipole's location: With regard to the scalp, a location in a gyrus will have a radial orientiation (1), while a location in a sulcus will result in a dipole orientation that is tangential to the scalp (2). In the latter case, the dipole moment will be perpendicular to the central sulcus as well as tangential to the scalp.
The figure displays typical mu rhythm scalp potential distributions (adapted from B Blankertz, R Tomioka, S Lemm, M Kawanabe, and KR Müller: Optimizing spatial filters for robust EEG single-trial analysis. IEEE Signal Proc. Magazine, 25(1):41-56, January 2008, reproduced with permission of the authors).
The distribution on the left illustrates the topography associated with a radially oriented source dipole (1) located on the right hemispheric motor gyrus. The distribution to the right is due to a tangentially oriented source dipole located in the central sulcus (2).
For source locations intermediate between (1) and (2), dipole orientation will be a linear combination of radial and tangential orientations. On the scalp, this results in a linear combination of associated scalp potential distributions (1) and (2).
The mu rhythm has an arc-shaped, periodic wave form (top left). In the frequency domain, such a waveform corresponds to a line spectrum with a strong first harmonic (bottom left). This means that there will be a second peak in the beta band, located at exactly twice the frequency of the first peak. Most often, relative modulation (i.e., change in amplitude relative to mean amplitude) is identical for both peaks (spectrum of actually measured signals to the right).
As discussed above, a human subject can wilfully influence the amplitude of her/his mu rhythm by imagination of hand or feet movement. Continuous feedback of mu rhythm amplitude can help improve this natural ability by selective reinforcement of successful strategies.
Much like a historic AM radio receiver, a mu rhythm BCI treats the mu rhythm as a carrier signal with information impressed on it by amplitude modulation. Consequently, its signal processing chain is analogous to that of an AM receiver.
Using a linear combination of simultaneous input samples, the spatial filtering step favors signals originating from hand/feet areas over signals that originate from other areas. In the AM receiver analogy, this step corresponds to a directional antenna that favors radio signals originating from the spatial direction corresponding to a desired broadcasting station's position over signals from undesired broadcasting stations, or spatially inhomogeneous noise.
All mu rhythm BCIs employ some type of frequency selection, favoring signals in a narrow band around a single peak, or multiple peaks, of the mu rhythm's spectrum. There is a number of possibilities to implement frequency selection; most common are
- IIR bandpass filtering,
- windowed spectral estimation methods such as
While IIR bandpass filtering is the direct computational analog of an AM receiver's tuning circuit, spectral estimation methods provide amplitudes for all frequency bands simultaneously, and require actual frequency selection in a separate classification step.
After extracting the carrier signal by spatial and temporal filtering, its amplitude time series (envelope) must be computed to obtain the original signal impressed onto the carrier. In a simple AM receiver, this is done using a rectifier diode in conjunction with a low pass circuit. If a BCI employs bandpass filtering for frequency selection, calculating the mean amplitude over a short interval usually performs the equivalent function.
For BCIs using spectral estimation methods, demodulation is often implemented inside the spectral estimation step, with its output being a distribution of absolute amplitude values.
As a next step, learn how to set up an EEG measurement.