12 - Variational mode decomposition-based finger flexion detection using ECoG signals

Abstract

The finger flexion movement prediction is a challenging problem of the brain-computer interface. This chapter focuses on decoding the finger flexion movement using electrocorticogram (ECoG) signals. The variational mode decomposition (VMD) is applied to obtain the subcomponents of each channel ECoG recording. Various correlation-based and other parameters such as correntropy, cross-information potential, and entropy estimation by Kozachenko-Leonenko are evaluated to categorize the flexion movement. The parameter computation over multiple channels provokes a complicated process. This complication is reduced by applying correlation-based thresholding between all channels, and significant channels are selected. All the computed features are given to the cubic support vector machine (C-SVM) classifier for classification. The complete model is investigated for the BCI competition-IV dataset, 2008, which holds brain signals of three subjects with different finger flexion movements. We have achieved the 0.43 correlation and 50.1% accuracy for five finger flexion classification on the considered dataset, and for index finger versus little finger flexion classification, 80.1% accuracy is achieved with 83% sensitivity.