DIPFIT

DIPFIT is a dipole fitting tool used in EEG and MEG analysis, commonly integrated with the EEGLAB environment. It models neural sources as equivalent current dipoles and estimates their location, orientation, and strength to explain measured scalp potentials or fields. DIPFIT is typically employed after artifact correction and, in many workflows, after independent component analysis to localize brain sources of components or ERP/ERSP data.

The method relies on forward models that relate a dipole inside the brain to the observed scalp activity. DIPFIT supports different head models, including spherical models and more realistic boundary element models (BEM), and requires electrode coordinates and a head model. For each chosen data unit (such as a component or an averaged epoch), the algorithm searches for a dipole that minimizes the residual variance between the observed data and the model prediction. Optimization methods commonly used include nonlinear least squares, such as Levenberg–Marquardt, to estimate the dipole’s position (x, y, z), orientation, and moment.

Output from DIPFIT typically includes the estimated dipole location, dipole moment, and a residual variance or goodness-of-fit metric. Visualization features allow users to display the dipole on a 3D head model and to inspect the fit relative to the data. DIPFIT outputs can be used to interpret the neural generators of ERP components or ICA-derived sources in anatomical terms, contributing to source localization analyses.

Limitations include reliance on the accuracy of the head model and electrode positions; dipole fitting assumes discrete sources and can be ill-posed when multiple sources contribute similarly to the data.