ConvDip: A Convolutional Neural Network for Better EEG Source Imaging

The electroencephalography (EEG) is a well-established non-invasive method in neuroscientific research and clinical diagnostics. It provides high temporal but low spatial resolution of brain activity. To gain insight about the dynamics EEG, one has to solve inverse problem, i.e., finding neural sources that give rise recorded EEG problem ill-posed, which means more than configuration can evoke same distribution activity on scalp. Artificial networks have been previously used successfully find either or two dipole sources. These approaches, however, never solved distributed model with We present ConvDip, novel convolutional network (CNN) architecture, solves based simulated data. show (1) ConvDip learned produce solutions from single time point data (2) outperforms state-of-the-art methods all focused performance measures. (3) flexible when dealing varying number sources, produces less ghost misses real comparison methods. plausible for recordings human participants. (4) trained needs <40 ms prediction. Our results qualify as an efficient easy-to-apply source localization data, relevance applications, e.g., epileptology real-time applications.