Neonatal seizure detection using atomic decomposition with a novel dictionary.

Area of Science:

• Signal Processing
• Biomedical Engineering
• Neonatal Neurology

Background:

• Atomic decomposition (AD) is an efficient signal decomposition technique.
• Neonatal seizures require accurate and timely detection for effective management.
• Existing methods for electroencephalogram (EEG) seizure detection can be complex.

Purpose of the Study:

• To develop and validate a novel method for neonatal seizure detection based on AD.
• To utilize a custom dictionary of pseudoperiodic Duffing oscillator atoms for improved seizure epoch coherence.
• To assess the efficacy of relative structural complexity as a sole feature for seizure detection.

Main Methods:

• A novel application-specific dictionary composed of pseudoperiodic Duffing oscillator atoms was designed.
• Orthogonal matching pursuit was employed for signal decomposition using the custom dictionary.
• Relative structural complexity, derived from AD convergence rate, was used as the seizure detection feature.

Main Results:

• The proposed method was tested on a large dataset of 826 hours of neonatal EEG data.
• The system demonstrated high performance, achieving a median receiver operator characteristic area of 0.91 (IQR 0.87-0.95) across 18 neonates.
• The relative structural complexity feature effectively distinguished seizure epochs from non-seizure periods.

Conclusions:

• The developed atomic decomposition method with a novel dictionary provides an effective approach for neonatal seizure detection.
• The proposed feature, relative structural complexity, is a promising sole indicator for identifying seizures in neonatal EEG.
• This technique holds potential for improving the clinical diagnosis and management of neonatal seizures.