Fetal weight estimation using the evolutionary fuzzy support vector regression for low-birth-weight fetuses

Area of Science:

• Medical Imaging
• Machine Learning in Healthcare
• Fetal Medicine

Background:

• Accurate fetal weight estimation is vital for managing low-birth-weight infants and preventing complications.
• Traditional regression analysis for fetal weight estimation using ultrasound measurements lacks sufficient accuracy, especially for low-birth-weight fetuses.
• Inter-observer variability and inappropriate regression models contribute to inaccuracies in current fetal weight estimation methods.

Purpose of the Study:

• To develop and validate a novel method, evolutionary fuzzy support vector regression (EFSVR), for improving fetal weight estimation accuracy in fetuses under 2500 g.
• To enhance the robustness of support vector regression (SVR) models against noisy data and measurement variability using fuzzy logic.
• To optimize the generalization performance of the fuzzy SVR (FSVR) model through the nondominated sorting genetic algorithm (NSGA).

Main Methods:

• Implementation of a novel EFSVR model integrating fuzzy logic with SVR and optimizing parameters using NSGA.
• Utilizing ultrasound measurements for fetal weight estimation.
• Comparative analysis against standard regression formulas, back-propagation neural networks, and basic SVR.

Main Results:

• The EFSVR model achieved the lowest mean absolute percent error (6.6%) and the highest correlation coefficient (0.902) for estimated fetal weight versus actual birth weight.
• EFSVR demonstrated significant accuracy improvements (1.9%-4.2%) over widely used fetal weight estimation formulas.
• The proposed model shows enhanced robustness to data variability compared to conventional methods.

Conclusions:

• The EFSVR model offers a significant advancement in accurate fetal weight estimation, particularly for low-birth-weight fetuses.
• This method shows considerable potential for improving clinical prenatal care and management of at-risk pregnancies.
• The integration of fuzzy logic and genetic algorithms enhances the reliability and accuracy of machine learning models in medical applications.