Multimodal Model for Predicting Fetal Acidosis in Delivery Room

Abstract

In the delivery room, fetal well-being is evaluated through laboratory tests, biosignals like cardiotocography, and imaging techniques such as fetal echocardiography. We have developed a multimodal machine learning model that integrates medical records, biosignals, and imaging data to predict fetal acidosis, using a dataset from a tertiary hospital's delivery room (n=2,266). To achieve this, features were extracted from unstructured data sources, including biosignals and imaging, and then merged with structured data from medical records. The concatenated vectors formed the basis for training a classifier to predict post-delivery fetal acidosis. Our model achieved an Area Under the Receiver Operating Characteristic curve (AUROC) of 0.752 on the test dataset, demonstrating the potential of multimodal models in predicting various fetal outcomes.