PROXIMAL GRADIENT-BASED STEP FOR BRAIN DISTRIBUTED EPILEPTIC SOURCE IMAGING AFTER TENSOR DECOMPOSITION PREPROCESSING
Abstract
Recently, a new two-step tensor-based distributed epileptic source localization method was proposed. Firstly, it performs Canonically Polyiadic (CP) decomposition of a Space-Time-Spike (STS) tensor capturing the occurrence of epileptic events in the ElectroEncephaloGraphic (EEG) data. This with the aim at estimating the spatial map of those distributed sources. Secondly, a source localization step is performed using the well-known Alternating Direction Method of Multipliers (ADMM) algorithm. This paper investigates to what extent this brain epileptic source imaging problem could be properly solved when a proximal gradient-based algorithm, the Proximal Alternating Linearized Minimization (PALM) technique, is used instead of the ADMM when the source localization step is considered. The convergence issue of the PALM scheme in this particular problem is discussed and a link with the one of the ADMM method is established. Furthermore, realistic simulations with epileptic EEG data show the efficiency of the new proposed PALM-based scheme.