Hidden Markov model and support vector machine based decoding of finger movements using electrocorticography

Abstract

Objective Support Vector Machines (SVM) have developed into a gold standard for accurate classification in Brain-Computer-Interfaces (BCI). The choice of the most appropriate classifier for a particular application depends on several characteristics in addition to decoding accuracy. Here we investigate the implementation of Hidden Markov Models (HMM)for online BCIs and discuss strategies to improve their performance. Approach We compare the SVM, serving as a reference, and HMMs for classifying discrete finger movements obtained from the Electrocorticograms of four subjects doing a finger tapping experiment. The classifier decisions are based on a subset of low-frequency time domain and high gamma oscillation features. Main results We show that decoding optimization between the two approaches is due to the way features are extracted and selected and less dependent on the classifier. An additional gain in HMM performance of up to 6% was obtained by introducing model constraints. Comparable accuracies of up to 90% were achieved with both SVM and HMM with the high gamma cortical response providing the most important decoding information for both techniques. Significance We discuss technical HMM characteristics and adaptations in the context of the presented data as well as for general BCI applications. Our findings suggest that HMMs and their characteristics are promising for efficient online brain-computer interfaces.

Extracting duration information in a picture category decoding task using hidden Markov Models

“…Consequently, the resulting features vary slowly over time (compare figure S1). Both feature extraction routines are described in detail in a previous study [14]. Channel selection is performed on training data only, using a two stage approach.…”

“…Channel selection is performed on training data only, using a two stage approach. First, an algorithm based on the Davies–Bouldin index [14] is applied to select a predefined number of most informative channels for the picture category separation problem. The corresponding labels of the training data are used to select the channels.…”

“…For initialization of the HMMs we reused the algorithm presented in [14]. The routine is based on k -means clustering with several small extensions that allow for adaptions to specific needs originating by the nature of the data.…”

The anterior contralateral response improves performance in a single trial auditory oddball BMI

“…SVMs are capable of delivering high and reproducible classification performance as well as robustness with a low number of training samples [27,28]. In this study, an SVM using a radial basis function kernel (RBF) was used to identify the attended target stimulus.…”

“…SVM has been shown to be a gold standard for accurate classification in BMI [28]. Guo et al have explored three classification methods to assess the target detection accuracy of auditory BMIs, i.e., area comparison, Fisher's discriminate analysis and SVM [44].…”