Model Misspecification Detection in Dipole Analysis

This project is intended to provide computational resources to users of Dipole Source Localization(DSL) methods to allow them to decide whether or not their DSL results are distorted by certain errorscalled model misspecification errors. These errors are essentially due to incorrect weighing constantsin the computation of the relationship between source dipole parameters and the surface potential. Sucherrors occur when head model simplifications are used, such as a one-shell as contrasted to a morerealistic 3-shell model, and when a spherical model is employed instead of a more realistic head model.Errors in single dipole localization are magnified when multiple dipoles are required to model the sourcesof brain potentials. A relatively simple method for estimating misspecification errors has been devised(called the AB-plane technique) and examined using simulated data for models using one or two dipoles.The method permits the identification of models that are insufficiently accurate to produce valid DSL,and to compare models for their relative accuracy. In order to use the method on real (as contrasted tosimulated) data in which the contributions of each dipole is unknown an extension of the AB-planeestimation technique called the EF-plane method will be used. Although accurate definition of the EFplane is not possible, since the variables that define it are unknown, it is proposed that a close enoughestimate might be obtained using singular value decomposition. It is indicated that even if this approachis not entirely accurate, that it will nevertheless be sufficiently close to the AB-plane estimate to disclosemodel misspecification errors.