Comparison of Z and multivariate statistical brain electromagnetic maps for the localization of brain lesions

Conventional Z maps provide probability statements about the deviation of observed values from the norm. Galán et al. (1994) introduced Simultaneous Significance Probability Scales to detect abnormalities over the whole map, making use of the information provided by the topographic structure of dependencies. They also described multivariate brain electromagnetic (MBE) maps for compact presentation of complex spatio-temporal information. In this paper, using the distance-based localization receiver operating characteristic curves (DL-ROC curves), we compare the localization provided by computed tomography with that provided by Z and MBE maps in 61 patients with brain lesions. Maps were calculated for absolute power and relative power in delta, theta, alpha and beta bands for voltage and current source densities (CSD). In each patient, all maps were compared and the map with the highest value of the area of the DL-ROC curve was considered to be the "best map." Z maps of CSD were the "best maps" in 24 patients. In the voltage montage, we observed that multivariate maps add some information not contained in the Z maps. However, for CSD, Z maps were more accurate than multivariate maps. A very consistent finding was the observation that lesions were better detected by maps analyzing the delta band, while edema was better represented by maps in the theta range.