A Deep Learning Approach to Grading Neonatal Hypoxic-Ischemic Encephalopathy Using ECG Spectrograms

Area of Science:

• Biomedical Engineering
• Neonatal Neurology
• Computational Medicine

Background:

• Hypoxic-Ischemic Encephalopathy (HIE) is a serious condition in newborns caused by oxygen deprivation to the brain.
• Accurate grading of HIE is crucial for determining eligibility for therapeutic hypothermia.
• Electroencephalogram (EEG) is currently used for HIE grading, but Electrocardiogram (ECG) signals are more accessible in Neonatal Units.

Purpose of the Study:

• To explore and compare two distinct methods for automatic HIE grading using neonatal ECG signals.
• To evaluate the efficacy of a novel deep learning approach utilizing ECG spectrograms against a classical feature extraction method.
• To establish ECG as a viable and accessible alternative to EEG for HIE assessment.

Main Methods:

• A large dataset of newborn ECG signals was utilized.
• Classical approach: Heart Rate (HR) calculation, time/frequency domain feature extraction, and classification (Random Forest, multilayer perceptron).
• Proposed approach: Direct extraction of spectrograms from ECG signals for input into a convolutional neural network.

Main Results:

• The proposed deep learning method achieved a higher Area Under the Curve (AUC) compared to the classical approach.
• The convolutional neural network approach bypassed the need for time-consuming HR calculation.
• Deep learning models demonstrated greater robustness in grading HIE from ECG data.

Conclusions:

• ECG signals can be effectively used for automatic HIE grading, offering a more accessible alternative to EEG.
• The deep learning approach using ECG spectrograms is a promising and efficient method for HIE assessment.
• This technique may enhance clinical decision-making regarding therapeutic hypothermia for neonates with HIE.