Combining EEG and MRI in comatose cardiac arrest patients

Background: Prognosis of comatose patients after cardiac arrest remains uncertain in ±50%. Combining data from different modalities can be used to enhance understanding of the comatose brain and, in term, improve outcome prediction. In this study, we combine EEG and MRI to find intermodal relationships and enhance understanding of the brain after cardiac arrest. Methods: Continuous EEG of comatose patients after cardiac arrest was recorded during the first 3 days and visually classified as either continuous, discontinuous, epileptiform or synchronous burst-suppression (BS). An MRI scan was made on day 3±1 after cardiac arrest and visually scored for ischemic damage in 21 brain regions on DWI and T2 FLAIR. Prominence of cortical and deep medullary veins and the number of microbleeds were scored on VenBold or SWI. MRI scores were related to the EEG categories to assess intermodal relationships. Quantitative analysis of EEG dynamics at 24h was performed using a model discovery algorithm called SINDy. The fit of the model, complexity of equations, local equilibria, autocorrelation-, interaction- and bias fraction were assessed per EEG category. Results: We included 75 and 67 patients for visual and quantitative analysis, respectively. Patients exhibiting epileptiform and synchronous BS patterns showed more abnormalities on DWI and T2 FLAIR than patients with (dis)continuous patterns. Ischemic damage was most prominent in cortical grey matter. Prominence of cortical and deep medullary veins was lower in patients with a synchronous BS pattern compared to patients with a continuous pattern. Quantitative features of EEG dynamics did not show significant differences between EEG categories. Conclusion: EEG patterns are related to abnormalities on MRI. Patients exhibiting EEG patterns associated with poor outcome have more abnormalities on DWI and T2 FLAIR than patients with EEG patterns associated with good and undetermined outcome. Multimodal analysis of brain dynamics with SINDy was not feasible using the current methodology.