Boosting-LDA algriothm with multi-domain feature fusion for motor imagery EEG decoding

Stroke results in uncoordinated limb movements of patients, greatly affecting their quality of life. Deep participation of patients with stroke in active rehabilitation training using motor imagery electroencephalogram (EEG) signals can greatly improve the rehabilitation efficiency. At present, the brain-computer interface (BCI) based on motor imagery is mostly in the laboratory research stage, and the participants are mostly healthy people. Understanding EEG differences between healthy people and stroke patients is important. A Novel EEG decoding algorithm is proposed based on this present situation, which adopt the strategy of multi-domain feature complementary fusion in feature extraction, and the ensemble learning to enhance the robustness of the model. Multi-scale features were extracted from time domain, frequency domain, space domain and time-frequency domain for fusion, to effectively utilize them to improve the classification accuracy. The ensemble linear discriminant analysis (LDA) classifier based on Boosting algorithm is proposed to make boosting in the multidomain feature level, and extract and optimize the most discriminative features from the high-dimensional feature combination space, which maximizes the ratio of the discreteness between inter-class and intra-class. Then, the public dataset and the collected EEG dataset of healthy subjects and stroke patients are used to validated the effective of proposed algorithm, and neural activation characteristics of participants during motor imagery processing are analyzed. Compared with the single feature classification algorithm, the proposed method has better positive effects on classification accuracy, sensitivity, specificity, and Kappa, which opens up new possibilities for the usage of brain-controlled active rehabilitation devices.

Automated summary

A novel EEG decoding algorithm based on motor imagery is proposed, utilizing a multi-domain feature complementary fusion strategy and ensemble learning to enhance model robustness, which effectively improves rehabilitation training efficiency and accuracy for stroke patients.