Advancing Fetal Surveillance with Physiological Sensing: Detecting Hypoxia in Fetal Sheep

Accurate detection of hypoxia during fetal monitoring is critical for timely intervention and prevention of brain injury. This study investigated hypoxia detection in fetal sheep by developing a classifier to differentiate between baseline, during acute hypoxia (umbilical cord occlusion; UCO), and post-hypoxia recovery (post-UCO), using physiological signals and an XGBoost machine learning model. A multi-modal approach integrating electrocardiogram (ECG), electroencephalogram (EEG), and electromyogram (EMG) signals was compared to a model based on the ECG signal only. We included recordings from 14 fetal sheep, each of which underwent acute hypoxia via a single UCO procedure. Performance of hypoxia detection was measured by Marco F1 score (F1 m), Accuracy (Acc), and standard deviation of validation error (SDE). Our multi-modal model achieved F1m= 0.88, Acc = 0.90, while the ECG-based approach achieved F1m = 0.82, Acc = 0.85. These results indicate that the multi-modal approach was superior to the ECG-only model for overall classification. While ECG and multimodal models correctly classified over 92% and 97% of acute hypoxia intervals, respectively, the classification of post- UCO intervals proved more challenging. The complexity of interpreting physiological signals underscores the need to consider temporal dynamics and physiological context in fetal monitoring.