Motor Imagery

Project to analyze the accuracy of multiple algorithms published in papers about EEG binary motor imagery problem.

The motor imagery (MI) based BCI is capable of translating the Subject’s movement intention to controls the external devices. Imagining a movement or performing an action mentally is known as MI. In MI tasks, subjects are instructed to imagine themselves performing a specific motor action (e.g. hand, foot) without overt motor output and each task is treated as a MI class.

Notes

• Was noted that in the covariance matrices estimation in the CSP algorithm, is better to use metric logeuclid or euclid instead riemman, because the accuracy is practically the same, and the riemman is much slower.

Results

FBCSP

BCI competition III dataset IV-a

> Accuracies obtained in reproduction of the article in fbcsp folder
GNB: 87.3366 +/- 9.3793
SVM: 87.1405 +/- 9.9846
LDA: 86.3889 +/- 10.1286

> Accuracies obtained applying manual selection of 8 electrodes in motor cortex area
GNB: 80.8725 +/- 12.0977
SVM: 81.9150 +/- 10.2631
LDA: 82.0229 +/- 10.4780

SI-BCI

BCI competition III dataset IV-a

> Accuracies obtained in reproduction of the article in si_bci folder
SVM
    CSP: 88.1429 +/- 11.1767
    KATZ_FRACTAL: 90.2143 +/- 8.3989
LDA
    CSP: 87.5714 +/- 10.7033
    KATZ_FRACTAL: 90.8571 +/- 7.6884

> Accuracies obtained applying manual selection of 8 electrodes in motor cortex area
SVM
    CSP: 81.2857 +/- 11.9493
    KATZ_FRACTAL: 81.4286 +/- 11.9352
LDA
    CSP: 80.2857 +/- 11.1584
    KATZ_FRACTAL: 81.6429 +/- 9.3503

SPECIFIC-BAND-CSP-FEATURES

BCI competition III dataset IV-a

> Accuracies obtained in reproduction of the article in specific_band_csp_features folder
GNB: 81.9286 +/- 11.4377
SVM: 87.9286 +/- 8.0143
LDA: 86.6429 +/- 9.8408

> Accuracies obtained applying manual selection of 8 electrodes in motor cortex area
GNB: 76.1429 +/- 11.8330
SVM: 81.7143 +/- 14.7365
LDA: 83.0714 +/- 12.5318