Machine learning ensemble modelling to classify caesarean section and vaginal delivery types using Cardiotocography...

Area of Science:

• Medical Informatics
• Signal Processing
• Machine Learning

Background:

• Cardiotocography (CTG) trace interpretation is crucial for fetal monitoring during labor.
• High inter- and intra-observer variability in visual interpretation leads to misinterpretations and adverse perinatal outcomes.
• Machine learning offers a potential objective decision support system for obstetricians and midwives.

Purpose of the Study:

• To review human Cardiotocography trace interpretation.
• To propose and evaluate a machine learning-based decision support system for objective CTG analysis.
• To enhance predictive capacity and reduce negative outcomes in fetal monitoring.

Main Methods:

• Feature set engineering using an open database of 552 intrapartum CTG recordings.
• Application of signal processing techniques to extract 13 features from raw CTG fetal heart rate traces.
• Recursive Feature Elimination for feature selection, addressing class imbalance with Synthetic Minority Oversampling Technique (SMOTE).
• Training and evaluation of machine learning algorithms, including an ensemble classifier (Fisher's Linear Discriminant Analysis, Random Forest, Support Vector Machine).

Main Results:

• An ensemble classifier achieved 87% Sensitivity (95% CI: 86%, 88%) and 90% Specificity (95% CI: 89%, 91%).
• The model demonstrated a high Area Under the Curve (AUC) of 96% (95% CI: 96%, 97%).
• A Mean Square Error (MSE) of 9% (95% CI: 9%, 10%) was recorded, indicating robust performance.

Conclusions:

• Machine learning, as a decision support tool, can provide an objective measure for Cardiotocography interpretation.
• The proposed system shows significant potential to increase predictive accuracy in fetal monitoring.
• Implementing such systems can help reduce adverse perinatal outcomes and improve patient safety during labor.