Solving the Inverse Problem for Auditory Steady-State Response via Sparse Bayesian Learning

Auditory steady-state response (ASSR), as a kind of steady-state EEG signal evoked by periodic auditory stimulation, has shown its potential for clinical application. To illustrate the rationality of its clinical application, it is important to study its generation mechanism, with the aid of ASSR source localization results. As the source activities of ASSR have the sinusoidal form, it would be simpler and more robust to consider the source localization problem based on the Fourier component of EEG recordings. However, there is currently no source localization algorithm specially designed for ASSR. In this paper, based on the empirical Bayesian framework, the ASSR source localization problem was modeled on the Fourier component instead of the recorded data itself. In consideration of the joint sparsity among various data segments, a sparse Bayesian learning (SBL) algorithm was developed, realized by the idea of expectation maximization (EM). The localization accuracy of proposed SBL algorithm is verified by simulation experiments using synthesized ASSR data. The origins of 40-Hz ASSRs evoked by multiple types of auditory stimulation were explored by the proposed algorithm, where the EEG measurements were recorded from multiple subjects.