Comparison and improvements of LCMV and MUSIC source localization techniques for use in real clinical environments

► We present some improvements on SVD-based methods for source localization in MEG. ► We concentrate in practical implementations for use in MEG commercial equipment. ► These improvements are tested in a practical case of auditive stimulation. ► Some strategies are proposed that can be implemented in commercial equipments. ► We focus on real, clinical scenarios. The present work shows some improvements realized on practical aspects of the implementation of Singular Value Decomposition (SVD) methods to localize the sources of neural activity by means of magnetoencephalograph (MEG). Two methods have been improved and compared i.e. a spatial filter, the Linearly Constrained Minimum Variance Beamformer (LCMV) method, and a signal subspace method that is an implementation of the MUSIC (Multiple Signal Classification) method due to Mosher et al. (1992). It also shows the performance of both methods comparing three different averaging procedures. The influence of the correct selection of the noise subspace dimension has been analyzed. Using acoustic stimulus for real patient measurements, we discuss the relevant differences of both methods and propose an adequate strategy for future diagnosis based on correct source localization.