Maximum-likelihood detection and estimation performance for rank one MEG activity

We investigate the performance of maximum-likelihood (ML) signal estimation and the corresponding generalized likelihood ratio test (GLRT) detector for rank one and full rank signal models. Spatio-temporal prior information is incorporated by constraining the subspace in which the signal lies. The rank one model enables estimation of the signal spatial pattern, as well as the corresponding signal time series. We consider ML detection and estimation performance as a function of subspace dimension and various levels of prior information regarding the source spatial pattern.Simulated and measured evoked response data are used to illustrate the performance advantages of the rank one signal model. The receiver operating characteristic shows that the benefits of rank one detection are greatest at moderately low SNR, and increase as the dimension of the subspace describing the prior information increases. The ML signal estimate is evaluated in terms of mean squared error. The rank one signal estimate is less sensitive to noise than the corresponding full rank signal estimate.