Machine Learning in Neonatal Respiratory Distress Syndrome: From Early Warning to Clinical Translation

Neonatal respiratory distress syndrome (NRDS), driven by pulmonary surfactant deficiency in preterm infants, is a leading cause of neonatal morbidity and mortality. This narrative review synthesises machine learning (ML) applications across the NRDS spectrum—from prenatal risk stratification to long-term complication prediction—appraised against the TRIPOD+AI (2024) framework. Reported discrimination is high but heterogeneous (AUC 0.78–0.99); the most robust estimate is the prospective Korean Neonatal Network study of 13,087 very-low-birthweight infants across 76 hospitals (full-model AUROC 0.9187; testing 0.9216). Ensemble methods and deep-learning architectures tend to outperform logistic regression, often by modest margins. Emerging modalities—lung-ultrasound radiomics, cry-signal analysis (∼95% accuracy, controlled datasets), and ATR-FTIR spectroscopy—remain research-stage. Reporting deficits are near-universal: <20% report calibration, none conduct prospective fairness audits, <15% release code. Clinical translation requires federated-learning consortia, explainable AI, prospective external validation, and structured implementation science.