Integrated Deep Learning and Supervised Machine Learning Model for Predictive Fetal Monitoring

Area of Science:

• Obstetrics and Gynecology
• Fetal Monitoring
• Computational Medicine

Background:

• Metabolic acidosis is a significant cause of fetal mortality.
• Current methods for assessing fetal acidosis have limitations, with many acidotic fetuses misclassified into lower-risk categories.
• Accurate prediction of fetal acidosis is crucial for timely intervention and improved outcomes.

Purpose of the Study:

• To develop a feature extraction and prediction algorithm for identifying fetal acidosis.
• To predict umbilical cord pH levels using fetal heart rate and uterine activity data.
• To create a robust tool for predictive fetal monitoring to assist obstetricians.

Main Methods:

• Developed feature extraction algorithms to identify key features from cardiotocography (CTG) data, including late and variable decelerations.
• Utilized an ensemble classification algorithm with 85% test accuracy for predicting fetal acidosis.
• Integrated a deep learning forecasting model (long short-term memory network) to predict fetal heart rate and uterine contractions.

Main Results:

• The developed algorithms accurately predict cord pH levels, a direct indicator of acidosis.
• The prediction system outperforms traditional category-based methods in identifying acidotic fetuses.
• The hybrid model effectively identifies fetal acidosis 2-4 minutes in advance.

Conclusions:

• The proposed deep learning and classification hybrid model provides a robust tool for predictive fetal monitoring.
• This methodology enables obstetricians to better assess fetal well-being and plan interventions.
• Early prediction of fetal acidosis using CTG data can significantly improve fetal outcomes.