| Abstract: | Power spectral and discriminant analysis techniques were used to compare EEG records obtained at term and at 3 months past term from 5 groups of varying risk and developmental outcome. The groups were: healthy full-terms; healthy pre-terms with normal outcomes; sick pre-terms with normal outcomes; sick pre-terms with delayed development; sick pre-terms with later neurological problems. The EEG samples recorded at term were identified as belonging to the correct subject group at 52-70% accuracy, 20% being chance for 5 groups. The accuracy varied with the 4 classes of EEG patterns used. The individual subjects were also classified into their correct groups with few exceptions. Similar success was obtained with EEG samples selected from recording at 3 months past term. The predominant power spectral discriminating features were changes in intra- and inter-hemispheric coherence, and increased power, particularly in the middle and higher frequency range. Thus, computer analyses of EEG samples, using features not readily identified visually, differentiated risk from non-risk infants and also differentiated infants with substantial neonatal medical complications who have good or poor developmental outcomes. |
