Artificial intelligence for preterm brith prediction and management: review

Background: Preterm birth is a leading cause of neonatal morbidity and mortality worldwide, affecting approximately ۱۰% of live births annually. Existing screening methods, including cervical length measurement and biochemical markers, have limited predictive accuracy. Artificial intelligence (AI) and machine learning (ML) have emerged as promising approaches for improving risk prediction by integrating complex clinical, imaging, and biological data. Objective: To review and critically evaluate current evidence on AI-based approaches for preterm birth prediction and management, with a focus on machine learning algorithms, feature selection methods, ensemble learning, explainable AI (XAI), and clinical applicability. Methods: A systematic review was conducted in accordance with PRISMA ۲۰۲۰ guidelines. PubMed, Web of Science, and Scopus were searched for studies published between January ۲۰۱۰ and January ۲۰۲۵. Eligible studies evaluated AI, ML, or deep learning models for predicting spontaneous preterm birth and reported quantitative performance metrics. Risk of bias was assessed using the PROBAST tool. Results: Twenty-two studies met the inclusion criteria. AI models were developed using electronic health records, cervical ultrasound imaging, radiomics, elastography, electrohysterography, and multi-omics data. Predictive performance varied considerably, with area under the receiver operating characteristic curve (AUC) values ranging from ۰.۶۲ to ۰.۹۸. Electrohysterography-based models achieved the highest discrimination, while transformer-based multi-omics models reported AUC values up to ۰.۸۹. Ensemble learning approaches, particularly stacking and blending frameworks, generally outperformed single-model strategies. However, most studies demonstrated a high risk of bias, limited external validation, inadequate calibration assessment, and poor reporting quality. The use of XAI and systematic feature selection methods remained limited. Conclusion: AI shows substantial potential to improve preterm birth prediction beyond conventional screening approaches. Future research should prioritize interpretable ensemble models, robust external validation, standardized reporting, and equitable implementation to support safe and effective integration into clinical practice.

کلیدواژه ها:

Preterm birth ، Artificial intelligence ، Machine learning ، Deep learning ، Explainable AI ، Ensemble learning ، Prediction models