Automatic alignment of EEG/MEG and MRI data sets

Objectives: We developed a new technique of fully automatic alignment of br ain data acquired with scalp sensors (e.g. electroencephalography/evoked po tential (EP) electrodes, magnetoencephalography sensors) with a magnetic re sonance imaging (MRI) volume of the head. Methods: The method uses geometrical features (two sets of head points: dig itized from the subject and extracted from MRI) to guide the alignment. It combines matching on 3 dimensional (3D) geometrical moments that perform th e initial alignment, and 3D distance-based alignment that provides the fina l tuning. To reduce errors of the initial guessed computation resulting fro m digitization. of the head surface points we introduced weights to compute geometrical moments, and a procedure to remove outliers to eliminate incor rectly digitized points. Results: The method was tested on simulated (Monte Carlo trials) and on rea l data sets. The simulations demonstrated that for the number of test point s within the range of 0.1-1% of the total number of head surface points and for the digitization error in the range of -2-2 mm the average map error w as between 0.7 and 2.1 mm. The average distance error was less than I mm. T ests on real data gave the average distance error between 2.1 and 2.5 mm. Conclusions: The developed technique is fast, robust and comfortable for th e patient and for medical personnel. It registers scalp sensor positions wi th MRI head volume with accuracy that is satisfactory for localization of b iological processes examined with a commonly used number of scalp sensors ( 32, 64, or 128). (C) 2001 Elsevier Science Ireland Ltd. All rights reserved .