MEG Source Localization Using Deep Prior

Rio Yamana, Hajime Yano, Ryoichi Takashima, Tetsuya Takiguchi, Seiji Nakagawa

Abstract: Estimating neuronal current sources in the brain from magnetoencephalography (MEG) or electroencephalography (EEG) is generally an underdetermined problem. Many conventional methods uniquely estimate the current source by explicitly assigning a prior distribution of current sources. However, the probability distribution of the actual current sources is not clear. In this work, the current source in the brain is estimated using an implicit prior distribution expressed by deep convolutional networks (deep prior), instead of the explicit prior distribution. As a result of performing current source estimation on simulated MEG data, it was shown that convolutional networks can express the prior distribution of the current sources.