Fetal Hypoxia Classification from Cardiotocography Signals Using Instantaneous Frequency and Common Spatial Pattern

Fetal hypoxia is a condition that is caused by insufficient oxygen supply to the fetus and poses serious risks, including abnormalities, birth defects, and potential mortality. Cardiotocography (CTG) monitoring is commonly used to identify fetal distress, including hypoxia, by categorizing cases as normal or hypoxia. However, traditional CTG interpretation, usually performed visually by experts, can be subjective and error-prone, resulting in observer variability and inconsistent outcomes. It highlights the need for an automated and objective diagnostic system to assist clinicians in interpreting CTG data more accurately and consistently. In this research, a fetal hypoxia diagnosis system is proposed based on CTG signals. The CTG dataset is first transformed into the time-frequency domain using instantaneous frequency and using common spatial pattern (CSP) for feature extraction. Finally, the extracted features are then used to train and evaluate four machine learning models for classification with a cross-validation 5-fold methodology. Objective criteria (pH values, BDecf, Apgar 1, and Apgar 5) and expert voting as a subjective criterion were used to classify the fetus as normal or hypoxia. The SVM model outperformed other models in detecting fetal hypoxia, achieving high accuracy across pH, BDecf, Apgar1, Apgar5, and expert voting in all steps. It achieved over 98% accuracy across all objective criteria and steps.