Combining State-Space Modeling, Biophysics, and Compressive Sensing to Solve Ill-Posed Inverse Problems for EEG and MEG Recordings

• Activity Number: 110
• Type: Invited
• Date/Time: Monday, August 10, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Imaging
• Author(s): Patrick L. Purdon* and Emery N. Brown
• Companies: Massachusetts General Hospital and Massachusetts General Hospital/MIT
• Keywords: EEG ; MEG ; source localization ; compressive sensing ; state space modeling

Abstract

Recording brain activity using high-density scalp electroencephalogram (EEG) or high-density magnetoencephalgraphy (MEG) to conduct behavioral and neurophysiological studies is becoming more commonplace. Indeed, recording simultaneous activity from 256 channels of EEG or from 306 channels of MEG along with 72 channels of EEG is routine in many institutions. Analyses of these high-dimensional, non-stationary time series are most frequently carried out in the recording space. However, in many instances, structural magnetic resonance imaging scans are conducted as part of the EEG or MEG/EEG study to enable localization of the sources within the brain responsible for the activity observed in the scalp recordings. The source localization problem is a high-dimensional ill-posed inverse problem as the number of recording sites is on the order of 102 whereas the number of sources is on the order of 105. Three approaches have developed to increase the accuracy of the solution to this inverse problem: using temporal constraints implemented through state-space modeling; using anatomic and neurophysiological constraints based on specific hypotheses derived from the study being conducted; and