Complexity-based graph convolutional neural network for epilepsy diagnosis in normal, acute, and chronic stages

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computational Biology

Background:

• Electroencephalography (EEG) is crucial for epilepsy diagnosis and management.
• Current EEG analysis effectively identifies normal and acute epilepsy phases.
• Distinguishing chronic epilepsy phases from normal states using EEG remains challenging.

Purpose of the Study:

• To investigate EEG signal complexity for characterizing normal, acute, and chronic epilepsy phases.
• To develop and validate an advanced epilepsy detection framework using EEG complexity features.

Main Methods:

• Computed five complexity indicators (approximate entropy, sample entropy, permutation entropy, fuzzy entropy, Kolmogorov complexity) from rat hippocampal EEG.
• Applied one-way ANOVA and principal component analysis to assess complexity features.
• Developed a graph convolutional neural network (GCNN) model utilizing multi-channel EEG complexity for phase classification.

Main Results:

• Complexity features successfully differentiated between normal, acute, and chronic epilepsy phases.
• The GCNN model achieved high prediction accuracy (>98%) and F1 scores (>97%) in classifying these three phases.
• EEG complexity characteristics proved significant for recognizing different epilepsy stages.

Conclusions:

• EEG complexity analysis offers a robust method for identifying distinct epilepsy phases.
• The proposed GCNN framework demonstrates superior performance in epilepsy phase detection.
• This research highlights the clinical significance of EEG complexity in epilepsy management.