Exploring Random Forest Machine Learning for Fetal Movement Detection using Abdominal Acceleration and Angular Rate...

Area of Science:

• Biomedical Engineering
• Maternal-Fetal Medicine
• Signal Processing

Background:

• Reduced fetal movement is a critical indicator of potential adverse perinatal outcomes.
• Improved fetal movement monitoring is essential for enhanced clinical decision-making.
• Wearable sensors and machine learning (ML) offer promising avenues for accurate fetal movement detection.

Purpose of the Study:

• To train and validate ML models for fetal movement detection.
• To explore the utility of angular rate data, in addition to acceleration, for ML model training.
• To compare the performance of ML models using acceleration features, angular rate features, and combined features.

Main Methods:

• Ten pregnant participants were enrolled in the study.
• Four abdominal inertial measurement units (IMUs) and one chest reference sensor were utilized.
• Three random forest classifiers were trained using acceleration, angular rate, and combined feature sets.

Main Results:

• All trained ML models demonstrated good classification performance, with Area Under the Receiver Operating Characteristic Curve (AUROC) ranging from 0.70 to 0.77.
• The model incorporating both acceleration and angular rate features exhibited a superior Positive Predictive Value (PPV) compared to models using single feature sets.
• This suggests that the combined feature set possesses enhanced discriminative power for fetal movement detection.

Conclusions:

• ML models trained on wearable sensor data can effectively detect fetal movements.
• Integrating angular rate data alongside acceleration data improves the accuracy and reliability of fetal movement detection.
• This approach holds potential for advancing fetal wellbeing monitoring and clinical decision support.