Multimodal Early Birth Weight Prediction Using Multiple Kernel Learning

Area of Science:

• Perinatal Medicine
• Machine Learning
• Biomedical Informatics

Background:

• Fetal weight is a critical indicator of fetal health.
• Early prediction of birth weight is essential for timely interventions.
• Multimodal data integration offers potential for improved predictive accuracy.

Purpose of the Study:

• To develop and validate a novel multimodal learning approach for early birth weight prediction.
• To utilize maternal-fetal variables from the first trimester of gestation.
• To enhance the assessment and monitoring of fetal health status.

Main Methods:

• Optimal selection of multimodal features using an ensemble-based approach.
• Application of a nonparametric Multiple Kernel Learning (MKL) regression algorithm.
• Kernel selection and weighting to maximize prediction performance.

Main Results:

• The proposed methodology achieved an absolute error of 234 g in birth weight prediction.
• Validated against state-of-the-art computational learning algorithms.
• Demonstrated the effectiveness of the multimodal feature selection and MKL approach.

Conclusions:

• The developed multimodal learning approach shows promise for early birth weight prediction.
• This method can serve as a valuable tool for early evaluation and monitoring of fetal health.
• Integration of diverse maternal-fetal data improves predictive capabilities for perinatal outcomes.