A Multimodal & Explainable Neuro-Boosting fusion Framework for Maternal Mortality and Morbidity

Abstract

Maternal deaths and pregnancy-related complications remain pressing global health concerns, especially in developing nations, where the ability to recognize high-risk pregnancies at an early stage plays a decisive role in clinical outcomes. Conventional risk evaluation techniques generally depend only on structured clinical measurements, which restricts their capacity to uncover complex and multidimensional patterns embedded in maternal health information. To overcome this shortcoming, the present research introduces an innovative hybrid framework named the Neuro-Boosting Fusion Network (NBFN), which combines deep learning with gradient boosting to enhance the accuracy of maternal risk estimation. The proposed system utilizes a Keras-based Convolutional Neural Network (CNN) to derive high-level visual descriptors from medical images and a Dense Neural Network (DNN) to represent structured clinical attributes such as maternal age, blood pressure readings, haemoglobin concentration, gestational duration, and obstetric history. These extracted representations are subsequently merged and passed to an Extreme Gradient Boosting (XGBoost) classifier for final decision-making. Clinical records were retrieved from the UCI Maternal Health Risk Dataset, whereas ultrasound imagery was obtained from openly accessible Kaggle repositories. Experimental evaluation revealed that the NBFN attained an accuracy of 95.8%, recall of 94.9%, and an AUC of 0.97, surpassing all comparative baseline methods. The framework represents a scalable, interpretable, and clinically viable decision-support system aimed at strengthening maternal healthcare outcomes.