Fetal Well-Being Prediction with One-Class Gaussian Process Anomaly Detection

Fetal heart rate (FHR) monitoring is vital to assess fetal well-being during labor. However, clinical decisions based on subjective visual interpretations can lead to inconsistencies, unnecessary cesarean sections, and legal disputes. The key challenges in the computerized analysis of FHR include class imbalance, where healthy cases vastly outnumber distress cases, lack of confidence score, as most approaches focus on classification rather than continuous fetal health assessment, and limited feature interpretability, which hinders clinical adoption. To address these challenges, we propose a One-Class Gaussian Process model trained on interpretable features from healthy FHR segments. This model learns the healthy FHR distribution and identifies potential anomalies. We further introduce the health confidence score (HCS), a continuous metric quantifying fetal well-being. This score offers clinicians an intuitive and interpretable measure of the fetus’s condition, thereby supporting timely and informed clinical decision-making. The results demonstrate the model’s robust 96% accuracy in classifying FHR segments.