Small-world characteristics of EEG patterns in post-anoxic encephalopathy

Abstract

Post-anoxic encephalopathy (PAE) has a heterogenous outcome which is difficult to predict. At present, it is possible to predict poor outcome using somatosensory evoked potentials in only a minority of the patients at an early stage. In addition, it remains difficult to predict good outcome at an early stage. Network architecture, as can be quantified with continuous electroencephalography (cEEG), may serve as a candidate measure for predicting neurological outcome. Here, we explore whether cEEG monitoring can be used to detect the integrity of neural network architecture in patients with PAE after cardiac arrest. From 56 patients with PAE treated with mild therapeutic hypothermia, 19-channel cEEG data were recorded starting as soon as possible after cardiac arrest. Adjacency matrices of shared frequencies between 1 and 25 Hz of the EEG channels were obtained using Fourier transformations. Number of network nodes and connections, clustering coefficient (C), average path length (L), and small-world index (SWI) were derived. Outcome was quantified by the best cerebral performance category (CPC)-score within 6 months. Compared to nonsurvivors, survivors showed significantly more nodes and connections. L was significantly higher and C and SWI were significantly lower in the survivor group than in the non-survivor group. The number of nodes, connections, and the L were negatively correlated with the CPC-score. C and SWI correlated positively with the CPC-score. The combination of number of nodes, connections, C, and L showed the most significant difference and correlation between survivors and non-survivors and CPC-score. Our data might implicate that nonsurvivors have insufficient distribution and differentiation of neural activity for regaining normal brain function. These network differences, already present during hypothermia, might be further developed as early prognostic markers. The predictive values are however still inferior to current practice parameters.

EEG as an Indicator of Cerebral Functioning in Postanoxic Coma

“…In particular, automated algorithms have been developed to reduce the subjective part of EEG interpretation (Noirhomme et al, 2014), but are still very far from routine clinical use. More advanced analyses, including single-trial topographic interpretation applied to mismatch negativity paradigms (Tzovara et al, 2013) or comparison of small-world characteristics of EEG spontaneous activity (Beudel et al, 2014) are showing promising results in outcome prediction, but, again, these are still the subject of scientific research and are not (yet) ready for clinical application. The increasing awareness of the ACNS nomenclature should lead to a uniform way of reporting EEG, and thus help with the cross-correlations of clinical reports (Sivaraju et al, in press;Westhall et al, 2015).…”

Electroencephalographic Findings in Posthypoxic Myoclonus

“…10 In the current Dutch guideline, 10 early prognostication after CPR mainly relies on somatosensory-evoked potentials (SSEPs) which only have a sensitivity for predicting bad outcome around 24%. 11 For this reason, new prognostication strategies, using more than 1 predictor, [12][13][14] are being developed. Since good early predictors of prognosis can both give more accurate information to the clinicians and the relatives of patients and reduce costs the of intensive care unit admission in case of an unfavorable prognosis, more reliable predictors are of great value.…”

Contribuciones de las técnicas machine learning a la cardiología. Predicción de reestenosis tras implante de stent coronario

“…La mayor parte de los esfuerzos se han enfocado en tareas de segmentación del tejido y estructuras anatómicas cardiacas (por ejemplo el endocardio), que por lo general es el paso previo para realizar cualquier otro tipo de estudio, como la detección de lesiones o la clasificación por patologías. El primer estudio puramente realizado con deep learning data de 2013, y en él se aplicó a la segmentación del ventrículo izquierdo a partir de ecocardiografía113 . Los concursos tipo datathon se han utilizado sistemáticamente desde 2009 como herramienta de explotación de imágenes cardiacas, principalmente con objetivos de segmentación del ventrículo izquierdo y derecho, a partir de diferentes imágenes, fundamentalmente resonancia magnética y tomografía14,100,114 .…”