Fetal birthweight prediction with measured data by a temporal machine learning method.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Obstetrics and Gynecology

Background:

• Birthweight is a critical indicator of fetal development and maternal-infant safety.
• Accurate birthweight measurement is challenging, often relying on imprecise empirical formulas.
• Existing estimation methods lack precision in clinical practice.

Purpose of the Study:

• To develop a hybrid birth weight prediction classifier using Long Short-Term Memory (LSTM) networks.
• To integrate electronic medical records and B-ultrasonic data for enhanced prediction accuracy.
• To compare the performance of the hybrid-LSTM model against other machine learning classifiers and empirical formulas.

Main Methods:

• Collected data from 5,759 Chinese pregnant women, including over 57,000 obstetric electronic medical records.
• Developed a hybrid classifier combining multiple electronic medical records and B-ultrasonic examination data.
• Utilized Long Short-Term Memory (LSTM) networks for temporal data analysis in prediction.

Main Results:

• The hybrid-LSTM model achieved the highest prediction accuracy (0.793 at 40th week) compared to CNN, RF, LR, SVR, and BPNN.
• The hybrid-LSTM model demonstrated superior accuracy and minimal Mean Relative Error (MRE) across different gestational periods.
• All machine learning models significantly outperformed traditional empirical formulas in birthweight prediction.

Conclusions:

• The developed hybrid-LSTM model offers a more accurate approach to birthweight prediction.
• Findings support the development of clinical delivery treatment guidelines and decision support systems.
• This model can assist clinicians in obstetric examinations and aid pregnant women in weight management.