Deep Learning–Driven Early Prediction of Bronchopulmonary Dysplasia Using Chest X-Rays and Clinical Data

Abstract

Early detection of Bronchopulmonary Dysplasia (BPD) is crucial for improving outcomes in preterm infants. This study presents a deep learning–based framework that combines chest X-ray imaging and clinical data for accurate BPD prediction. Raw chest X-rays undergoU-Net lung-field segmentation to isolate relevant regions, followed by feature extraction using a pretrainedResNet, generating robust image embeddings. These embeddings are integrated with clinical parameters such as gestational age and birth weight. An XGBoost classifier is then trained on the combined features to predict BPD risk. Evaluation demonstrates that the proposed framework achieves high sensitivity, specificity, and predictive accuracy, outperforming conventional clinical scoring methods. The approach offers a practical and interpretable tool for early BPD risk assessment, supporting timely clinical decision-making and intervention strategies. This work highlights the potential of AI-driven multimodal analysis for improving neonatal care.