Abstract
Hypoxic-ischemic encephalopathy (HIE) remains a critical challenge in neonatal intensive care, requiring precise bedside monitoring to mitigate long-term neurological morbidity. While individual assessment of cerebral and cardiovascular activity is standard, the dynamic interaction between these systems provides untapped diagnostic value. This pilot study investigates neuro-cardiac coupling using synchronized long-term EEG and ECG recordings, focusing primarily on a core subset of 61 neonates classified by clinical HIE severity (Sarnat Stage 1 vs. Stage 2). We employed cross-modal analysis to evaluate the relationship between the amplitude-integrated EEG (aEEG) envelope and heart rate variability (HRV) metrics. Our results demonstrate that neuro-cardiac interaction is significantly modulated by HIE severity and electrocortical background patterns. Neonates with healthy continuous normal voltage (CNV) patterns exhibited robust neuro-cardiac synchronization (median Spearman’s
r
=
0.22
), which was significantly reduced by approximately 60% in patients with abnormal EEG backgrounds (median Spearman’s
r
=
0.09
,
p
=
0.015
). To ensure robust translation, machine learning models were evaluated across diverse parameter grids. While isolated hyperparameter configurations yielded peak performance metrics–such as an overall maximum accuracy of 90.22% using combined cardiocycle and spectral ECG features with XGBoost–the feature space demonstrated strong, un-biased baseline stability, yielding reliable average accuracies across window lengths (e.g., 77.27% mean
F
1
-score for EEG-based Logistic Regression and 72.73% mean
F
1
for ECG-based XGBoost). These findings suggest that multi-modal coupling metrics offer enhanced physiological insight for automated HIE grading in a pilot setting.