[Paper Review] Bayesian Nonparametric Models for Synchronous Brain-Computer Interfaces

Jaime F. Delgado Saa, Müjdat Çetin|arXiv (Cornell University)|Dec 27, 2016

Bayesian Methods and Mixture Models10 references3 citations

TL;DR

This paper proposes a Bayesian nonparametric Hidden Markov Model (HDP-HMM) with sticky transitions for synchronous brain-computer interfaces, using hierarchical Dirichlet processes to infer the number of hidden states and Gaussian mixture components from data without prior specification. The method outperforms standard HMMs and the BCI Competition IV winner, achieving higher kappa values on EEG motor task classification by modeling temporal dynamics of EEG power in frequency bands.

ABSTRACT

A brain-computer interface (BCI) is a system that aims for establishing a non-muscular communication path for subjects who had suffer from a neurodegenerative disease. Many BCI systems make use of the phenomena of event-related synchronization and de-synchronization of brain waves as a main feature for classification of different cognitive tasks. However, the temporal dynamics of the electroencephalographic (EEG) signals contain additional information that can be incorporated into the inference engine in order to improve the performance of the BCIs. This information about the dynamics of the signals have been exploited previously in BCIs by means of generative and discriminative methods. In particular, hidden Markov models (HMMs) have been used in previous works. These methods have the disadvantage that the model parameters such as the number of hidden states and the number of Gaussian mixtures need to be fix "a priori". In this work, we propose a Bayesian nonparametric model for brain signal classification that does not require "a priori" selection of the number of hidden states and the number of Gaussian mixtures of a HMM. The results show that the proposed model outperform other methods based on HMM as well as the winner algorithm of the BCI competition IV.

Motivation & Objective

• To improve synchronous BCI performance by modeling the temporal dynamics of EEG signals during motor tasks.
• To overcome the limitation of traditional HMMs requiring pre-specified numbers of hidden states and Gaussian mixture components.
• To develop a Bayesian nonparametric approach that infers model complexity directly from EEG data.
• To enhance classification accuracy using time-frequency features and state dynamics in sensorimotor rhythm tasks.
• To provide a robust, data-driven alternative to fixed-parameter HMMs in BCI systems.

Proposed method

• Uses auto-regressive modeling to compute time-frequency distributions of EEG signals for feature extraction.
• Applies common spatial patterns (CSP) for spatial filtering to enhance signal-to-noise ratio.
• Employs a Hierarchical Dirichlet Process-Hidden Markov Model (HDP-HMM) to allow an infinite number of hidden states and Gaussian mixture components.
• Incorporates sticky transition priors to encourage state persistence and improve modeling of temporal sequences.
• Uses Bayesian inference to compute posterior distributions over model complexity, avoiding manual hyperparameter tuning.
• Applies the forward-backward algorithm for filtering and classification at the end of each trial.

Experimental results

Research questions

• RQ1 Can a Bayesian nonparametric model automatically infer the optimal number of hidden states in EEG-based BCI classification?
• RQ2 Does the use of hierarchical Dirichlet processes for both state and emission distributions improve classification performance over fixed-parameter HMMs?
• RQ3 How does the sticky HDP-HMM compare to the BCI Competition IV winner in terms of kappa score on the same EEG dataset?
• RQ4 To what extent do temporal dynamics of EEG power in frequency bands contribute to improved BCI performance?
• RQ5 Can the proposed model reduce reliance on cross-validation or expert tuning for HMM parameters?

Key findings

• The Sticky HDP-HMM achieved higher average kappa values than all other HMM-based methods and the BCI Competition IV winner, which reported a maximum kappa of 0.60.
• The proposed method outperformed the HMM with fixed parameters (HMM-FP), which used three states and two Gaussian components per state.
• The HMM with cross-validation (HMM-CV) selected up to three hidden states and three Gaussian components, but still underperformed the sticky HDP-HMM.
• The Bayesian nonparametric approach successfully inferred model complexity without requiring prior specification of the number of states or mixture components.
• The method effectively captured the temporal dynamics of EEG power changes in alpha and beta bands during motor imagery tasks.
• Artifact reduction via linear regression and CSP filtering improved signal quality, contributing to the overall performance gain.