Probabilistic graphical model identifies clusters of EEG patterns in recordings from neonates

Objectives: In this paper we introduce a novel method for the evaluation of neonatal brain function via multivariate EEG (electroencephalography) signal processing and embedding into a probabilistic graph, the so called Chow-Liu tree. Methods: Using 28 EEG recordings of preterm and term neonate infants the complex features of the EEG signals were constructed in the form of a Chow-Liu tree. The trees were embedded into a 3 dimensional Euclidean space. Clustering of specific EEG patterns was done by complete linkage algorithm. Results: Our analytic tool was able to build clusters of patients with pathological EEG findings. In particular, we were able to make a visual proof on a 3d multidimensional scaling coordinate system with a good performance. The distances (graph edit distance) between Chow-Liu trees of different infants were proportional to the clinical findings of corresponding infants. Conclusion: Our method may provide a basis for the future development of a diagnostic/prognostic non-invasive brain monitoring tool which will be able to differentiate between a variety of complex clinical findings. Significance: This model addresses relevant issues in neonatology and neuropediatrics in terms of identification of possible clinical factors which interfere with normal brain development and will allow fast unbiased recognition of infants with specific pathological EEG findings.