Effective classification for neonatal brain injury using EEG feature selection based on elastic net regression and...

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Pediatric Neurology

Background:

• Neonatal brain injury can lead to severe neurological deficits, including seizures and cerebral palsy.
• Electroencephalography (EEG) and machine learning are vital for assessing infant brain injury.
• Current methods using all EEG features can be computationally intensive and reduce classification performance.

Purpose of the Study:

• To develop a novel, efficient feature selection method for neonatal brain injury classification.
• To improve the accuracy and reduce the computational load of EEG-based brain injury assessment systems.

Main Methods:

• A hybrid feature selection approach, EN-ICSA, combining Elastic Net (EN) regression and an Improved Crow Search Algorithm (ICSA).
• EN regression performs initial feature pre-screening.
• ICSA utilizes dynamic perception probability, a novel neighbor-following strategy, and experience-based global search to optimize feature selection, avoiding local optima and enhancing search efficiency.

Main Results:

• The proposed EN-ICSA feature selection method, when integrated with a Support Vector Machine (SVM) classifier, achieved high performance.
• Achieved classification accuracy of 91.94%, precision of 92.32%, recall of 89.85%, and F1-score of 90.82%.
• Outperformed traditional machine learning and feature selection methods in neonatal brain injury assessment.

Conclusions:

• The EN-ICSA feature selection method significantly enhances the performance of EEG-based neonatal brain injury classification systems.
• This novel approach offers a more efficient and accurate tool for diagnosing and managing neonatal brain injuries.
• The findings suggest a promising direction for improving diagnostic tools in pediatric neurology.