Motor Imagery EEG Classification Based on Decision Tree Framework and Riemannian Geometry.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Motor imagery (MI) electroencephalography (EEG) classification is crucial for brain-computer interfaces (BCI).
• Existing methods struggle with nonseparable regions in Riemannian manifolds of covariance matrices.
• Improving classification accuracy and reducing data dimensionality are key challenges.

Purpose of the Study:

• To propose a novel classification framework and data reduction method for multiclass MI EEG.
• To enhance BCI performance by accurately distinguishing MI tasks.
• To leverage Riemannian geometry for robust EEG signal analysis.

Main Methods:

• Method 1: Subject-specific decision tree (SSDT) with filter geodesic minimum distance to Riemannian mean (FGMDRM) for MI task identification.
• Method 2: Combines semisupervised joint mutual information (semi-JMI) with general discriminate analysis (SJGDA) for feature extraction, and SSDT with k-nearest neighbor (SSDT-KNN) for classification.
• Utilizes the manifold of covariance matrices in a Riemannian perspective for signal processing.

Main Results:

• Method 2 achieved a kappa value improvement from 0.57 to 0.607 on the BCI competition IV dataset 2a, outperforming the previous winner.
• Method 2 demonstrated high recognition rates on two additional datasets.
• The proposed SJGDA feature extraction effectively reduces dimensionality in the Riemannian tangent plane.

Conclusions:

• The novel Riemannian-based classification framework and data reduction methods significantly improve MI EEG classification accuracy for BCIs.
• The SSDT-KNN approach with SJGDA offers a powerful solution for enhancing BCI performance.
• The findings contribute to advancing the field of BCI through advanced machine learning and signal processing techniques.