Artificial intelligence in cardiotocography: Prediction models and applications

Cardiotocography (CTG) is the most common method for intrapartum fetal surveillance, but its interpretation is limited by human variability and modest predictive value for adverse outcomes. Machine learning (ML) and deep learning (DL) techniques have been increasingly applied to CTG data in an eort to improve accuracy, consistency, and clinical utility. This article provides an overview of 20 peer-reviewed studies identied through structured searches in PubMed, IEEE Xplore, Scopus, and Google Scholar. The included work spans traditional ML approaches such as support vector machines, random forests, and ensemble methods, as well as advanced DL architectures including convolutional neural networks (CNNs), residual networks (ResNets), autoencoders, and hybrid 1D-CNN plus BiGRU models. Reported classication accuracies reached up to 99% and AUC values approached 0.99, particularly in binary classication tasks. However, performance often degraded for minority classes without explicit imbalance handling, and external validation was rare. Common datasets included CTU-UHB, Kaggle CTG, and the Porto database, with notable variation in preprocessing and evaluation strategies. Promising directions include real-time monitoring, rare pattern detection (e.g., increased FHR variability), and synthetic data generation using GANs. Despite strong retrospective performance, challenges remain in generalizability, interpretability, and integration into clinical workows. To support safe adoption, future 1 work should emphasize larger and more diverse datasets, standardized reporting, external validation, and explainable AI tools tailored for clinical decision-making.