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

Area of Science:

• Neuroscience
• Medical Informatics
• Signal Processing

Background:

• Neonatal brain function assessment is crucial for early detection of developmental issues.
• Current methods for analyzing neonatal electroencephalography (EEG) can be complex and time-consuming.
• There is a need for advanced tools to objectively evaluate brain activity in neonates.

Purpose of the Study:

• To introduce a novel method for evaluating neonatal brain function using multivariate EEG signal processing.
• To embed complex EEG features into a probabilistic graph structure, specifically a Chow-Liu tree.
• To develop a tool for identifying and differentiating pathological EEG findings in neonates.

Main Methods:

• Utilized 28 EEG recordings from preterm and term neonate infants.
• Constructed complex EEG signal features into Chow-Liu trees and embedded them in a 3D Euclidean space.
• Employed a complete linkage algorithm for clustering specific EEG patterns and used graph edit distance for comparison.

Main Results:

• Successfully built clusters of patients with pathological EEG findings.
• Visualized results using a 3D multidimensional scaling coordinate system, demonstrating good performance.
• Showed that distances between Chow-Liu trees were proportional to clinical findings in infants.

Conclusions:

• The developed method can form the basis for a future non-invasive diagnostic and prognostic brain monitoring tool for neonates.
• This approach facilitates the differentiation of various complex clinical findings based on EEG patterns.
• The model aids in recognizing infants with specific pathological EEG findings, addressing key issues in neonatology and neuropediatrics.