Machine Learning and Deep Learning in Detection of Neonatal Seizures: A Systematic Review

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Neonatal Neurology

Background:

• Neonatal seizures are common neurological emergencies requiring prompt detection.
• Machine learning (ML) and Deep Learning (DL) offer promising advancements for seizure management.

Purpose of the Study:

• To evaluate the impact of ML and DL techniques on the detection accuracy of neonatal seizures.
• To synthesize current evidence on ML/DL applications in neonatal seizure detection.

Main Methods:

• A comprehensive literature search was conducted across major databases (PubMed, Scopus, etc.) in April 2024.
• Keywords included "Neonatal," "seizure," "machine learning," and "detection."
• A scoping review process identified ten relevant studies from 3512 initial records.

Main Results:

• Analysis of 1389 seizures across 17-258 newborns in NICU settings.
• ML models demonstrated high performance: Area Under the Curve (AUC) ranged from 80.7% to 99.3%.
• Average sensitivity and specificity for seizure detection were between 60.4% and 93.38%.

Conclusions:

• Convolutional neural network models show significant potential for early and accurate neonatal seizure detection.
• Further development and validation of ML/DL models are recommended.
• Integration into neonatal intensive care units (NICUs) is crucial for clinical implementation.