A computational approach to using routine MRI and EEG to identify developmental impairment in pre-school children with epilepsy

Purpose: To apply a computational approach to identify children with early-onset epilepsy (CWEOE, onset <5 yrs.) with higher risk for cognitive and/or behavioural problems using routine MRI and EEG available from initial clinical presentation. 
Methods: MRI, EEG and psychometric data on 32 CWEOE from a regional population-based study (Hunter et al. Developmental Medicine & Child Neurology 2015;57:56-57.) were available for this cross-sectional study. Subcortical volumes, normalised by intracranial volume, were calculated from T1 MRI sequences. Twenty channels resting-state EEG were processed into a frequency spectrum between 0.5 and 45 Hz. Cognitive and behavioural scores were normalised to the same standard space. CWEOE were split into average (S+) and those with deficits (S-, <-2.0 standard deviation below the mean). MRIs, EEGs, and scores were jointly analysed by performing tensor-matrix-matrix decomposition simultaneously, a method to reveal underlying correlations among diverse datasets.
Results: The decomposition revealed four underlying components; one referring to S+, and another to S-. S+ and S- components were in the same EEG frequency range <10 Hz, with S- in the lower amplitude.The decomposition revealed asymmetry differences in the thalamus and hippocampus between S+ and S-. Asymmetry Index (AI) (Sarica et al. Frontiers in neuroscience 2018;12:576) was adopted to evaluate the result. ANOVA showed significant differences in thalamus-AI and behavioural scores (p=0.035), and hippocampus-AI and cognition/behaviour scores (p=0.039).The decomposition showed volume difference in several regions. At α=0.05, Kruskal-Wallis test confirmed that the left thalamus was significantly different in behavioural scores (p=0.025). Other regions showed a trend at α=0.1, which will be further explored by integrating DTI data into the decomposition in future.
Conclusion: Our novel computational approach using decomposition identified characteristic differences between S+ and S- using routine MRI and EEG data, can potentially help prioritise CWEOE who would benefit from developmental interventions.