Seizure Prediction Analysis of Infantile Spasms

Area of Science:

• Neurology
• Biomedical Engineering
• Computational Neuroscience

Background:

• Infantile spasms (IS) is a severe childhood epilepsy with significant adverse outcomes.
• Effective prediction of IS seizures is critical but understudied.
• Current prediction methods lack comprehensive analysis of IS mechanisms.

Purpose of the Study:

• To develop and validate a seizure prediction framework for infantile spasms.
• To combine statistical analysis of brain networks with deep learning models.
• To improve the accuracy and reliability of IS seizure prediction.

Main Methods:

• Scalp electroencephalograms (sEEG) data from 25 IS patients were analyzed.
• sEEG data were divided into five phases: Interictal, Preictal, SPH, Seizure, and Postictal.
• Brain network Phase-Locking Value (PLV) was constructed for five brain rhythms (θ, α, β, γ).
• Statistical analysis identified key brain regions and rhythms involved in seizure transitions.
• A Resnet18 deep learning model was employed for prediction and validation.

Main Results:

• Significant variability in prefrontal, occipital, and central brain regions during seizure transitions was observed.
• Theta (θ), alpha (α), beta (β), and gamma (γ) brain rhythms were predominant.
• The Resnet18 model achieved high performance: 79.78% accuracy, 94.46% specificity, and 75.46% recall.
• Statistical findings were consistently validated by the deep learning model.

Conclusions:

• The proposed framework effectively integrates statistical analysis and deep learning for IS seizure prediction.
• The study provides insights into the underlying mechanisms of IS seizure transitions.
• This work offers a guideline for developing intelligent, systematic models for comprehensive IS seizure prediction.