Motor imagery-based EEG signal processing

Motor Imagery (MI) focuses on analyzing and interpreting electroencephalographic (EEG) signals generated by the brain during the mental imagery of motor actions. This technique involves recording the electrical activity of the brain while a subject imagines performing specific motor tasks without actually physically moving.

The fundamental idea behind motor imagery-based EEG signal processing is that when a person imagines performing a certain motor action, the brain generates neural activity patterns that are somewhat similar to those generated during actual physical movement.

Motor Imagery Analysis Using EEG Frequency Bands:

• Sensorimotor Rhythm (SMR): SMR is a subset of the alpha band (8 - 13 Hz) that's prominent over the sensorimotor cortex. During motor imagery tasks, desynchronization (reduction in power) of the SMR is observed over the hemisphere corresponding to the imagined movement. For example, imagining moving the right hand leads to desynchronization in the left sensorimotor cortex.
• Beta Band: The beta band, especially the lower beta range (13 - 20 Hz), is associated with motor planning and execution. In motor imagery tasks, an increase in beta power is observed over the motor cortex corresponding to the imagined movement.
• Gamma Band: Gamma activity (30 - 100 Hz) is linked to complex cognitive functions and is involved in the binding of different brain regions. It can also be modulated during motor imagery tasks, particularly in tasks that involve detailed mental representations of movements.

Application in Brain-Computer Interfaces (BCIs) for Motor Rehabilitation:

• Stroke Rehabilitation: Individuals who have suffered from strokes or other motor impairments can use motor imagery-based BCIs to engage in virtual rehabilitation exercises. By imagining movements, they can control virtual avatars or robotic devices, promoting neural plasticity and functional recovery.
• Prosthetic Control: People with limb amputations can use motor imagery-based BCIs to control prosthetic limbs. By imagining the movement of their missing limb, they can trigger specific actions in the prosthetic device, allowing them to regain a degree of motor control.
• Cognitive Neuroscience and Understanding Brain Plasticity: Analyzing EEG frequency bands during motor imagery tasks offers insights into cognitive processes and neural plasticity:
• Neurofeedback: EEG-based neurofeedback allows individuals to learn how to modulate their brain activity. For motor imagery, participants can learn to increase or decrease specific frequency bands, potentially enhancing their motor imagery skills.

The process of motor imagery-based EEG signal processing typically involves several steps:

• Preprocessing: The raw EEG signals often contain various artifacts such as noise, eye blinks, muscle activity, and external interference. Preprocessing techniques are used to filter out these artifacts and enhance the quality of the EEG data.
• Feature Extraction: Relevant features are extracted from the preprocessed EEG signals. These features might include power spectral densities, event-related desynchronization/synchronization (ERD/ERS) patterns, or other time-frequency representations.
• Classification: Machine learning algorithms, such as support vector machines (SVM), artificial neural networks (ANN), or convolutional neural networks (CNN), are trained using the extracted features to classify EEG patterns into different motor imagery classes (e.g., left hand movement, right hand movement, etc.).
• Model Training: The classification model is trained using labeled data where the motor imagery tasks performed by the subject are known. The model learns to differentiate between different motor imagery classes based on the extracted features.
• Testing and Validation: The trained model is tested on new, unseen EEG data to evaluate its performance in accurately classifying motor imagery tasks. Cross-validation techniques are often used to assess the model's generalization ability.

---

Spatial Filters:
Spatial filters are used to emphasize certain spatial patterns of neural activity while attenuating noise or unwanted signals. They are commonly used for source localization, noise reduction, feature extraction, and improving the spatial resolution of EEG signals. Further information

• Common Average Reference (CAR)
• Principal Component Analysis (PCA)
• Independent Component Analysis (ICA)
• Minimum Norm Estimation (MNE)
• Laplacian Filter

---

Dataset
Parameters of data:

• cnt: the continuous EEG signals, size [time x channels]. The array is stored in datatype INT16. To convert it to uV values, use cnt= 0.1*double(cnt); in Matlab.
• mrk: structure of target cue information with fields (the file of evaluation data does not contain this variable).
  • pos: vector of positions of the cue in the EEG signals given in unit sample, length #cues.
  • y: vector of target classes (-1 for class one or 1 for class two), length #cues.
• nfo: structure providing additional information with fields
  • fs: sampling rate
  • clab: cell array of channel labels
  • classes: cell array of the names of the motor imagery classes
  • xpos: x-position of electrodes in a 2d-projection
  • ypos: y-position of electrodes in a 2d-projection. Objective: Detection (L) left hand, (R) right hand movement.

---

Outline of the method:

• Load Data
• Signal Preprocessing
• Feature Extraction
• Classification Model

Machine Learning for Classification (Further information)