Pregnancy monitoring and risk assessment
In the field of perinatal medicine, preterm birth, gestational diabetes mellitus (GDM) and pre-eclampsia (PE) are common complications that seriously affect maternal and infant health. AI technology can develop risk prediction models for these complications by deeply analysing multidimensional data (such as personal and family medical history, age, weight, blood pressure, blood sugar, lifestyle).7 8 Table 1 summarises key studies that have explored these applications of AI. Furthermore, AI demonstrates utility in predicting long-term outcomes such as the risk of postpartum progression to type 2 diabetes among patients with GDM.9
AI applications in pre-eclampsia prediction and management
Early screening and management of pre-eclampsia
PE is a severe disease unique to pregnancy and is one of the main causes of morbidity and mortality in pregnant women and perinatal infants. In recent years, AI technology has shown great potential in the prediction, diagnosis and management of PE [area under the curve (AUC)>0.9].10 For instance, the research team from Sun Yat-sen University employed AI deep learning algorithms to construct the PROMPT prediction model, which is based on retinal vascular features. This model achieved an AUC of up to 0.87 for predicting PE and an accuracy rate of up to 0.91 for predicting preterm PE.11 Additionally, machine learning models have significantly enhanced the accuracy of predicting PE by incorporating routine blood test data and clinical features. A model developed by a Brazilian study, which used synthetic data enhancement technology, achieved an accuracy rate of 90.6% (AUC=0.832) by considering parameters such as haemoglobin and liver function indicators.12 13 These technologies can identify high-risk pregnant women in advance and provide decision support for clinical doctors, thereby reducing the incidence of complications such as preterm birth and GDM, and ultimately improving pregnancy outcomes.
Value of artificial intelligence in labour prediction
Recent studies consistently indicate that AI in obstetrical decision support holds significant clinical potential for predicting the course of labour. Liu et al developed second-stage duration models using 1-hour and 2-hour thresholds. In the testing set, the 1-hour model achieved an AUC of 0.808, while the 2-hour model reached a higher AUC of 0.824, indicating strong discriminative performance for both.14 When dynamic intrapartum features such as the duration of the first stage, uterine contractions and fetal position are incorporated, the explanatory power for predicting the duration of the second stage is substantially enhanced.14 Fernández et al used a comprehensive clinical database containing 48 attributes to develop models predicting mode of delivery (caesarean section, eutocic vaginal delivery and instrumental vaginal delivery) with machine learning algorithm.15 In the caesarean versus vaginal delivery discrimination, all three models achieved overall accuracy at or above 90%, while discrimination between instrumental and vaginal deliveries approached 87%.15 Labour duration prediction tools offer potential clinical utility for real-time labour monitoring, determining whether to continue vaginal delivery or to initiate caesarean delivery earlier, and guiding resource allocation and delivery pathways. However, multiple studies emphasise the necessity of cross-institution external validation and local calibration to ensure reproducibility and safety across diverse patient populations and settings.14 16 17
Application of artificial intelligence in preterm birth
Preterm birth is one of the primary causes of perinatal infant mortality and long-term health issues globally, with an incidence rate of approximately 10%.18 In recent years, AI technology has demonstrated significant potential in predicting, preventing and managing preterm births. By analysing multidimensional data, (such as medical history, age, weight, lifestyle, genetic factors and biomarkers of pregnant women), AI can construct preterm birth risk prediction models. This enables the identification of high-risk pregnancies in advance, offering opportunities for early clinical intervention. For instance, researchers employed machine learning algorithms to analyse biomarkers, including inflammatory factors and hormone levels, in the blood samples of pregnant women. They combined this with clinical data, such as pregnancy history and pregnancy complications, to develop models capable of accurately predicting the risk of preterm birth.19 Additionally, AI technology can be used for real-time monitoring during pregnancy. By employing wearable devices and mobile medical technology, it can gather physiological data from pregnant women in real-time (including heart rate, blood pressure, uterine contraction frequency), analyse this data and promptly detect abnormal conditions, issuing alerts when necessary.20 As AI technology continues to evolve and data accumulates, its application in preventing and managing preterm birth is expected to expand and deepen. This development could significantly reduce the incidence of preterm births, enhance the prognosis for preterm infants and protect the health of both mothers and infants.
Prenatal ultrasound examination
Ultrasound examination is an important method for prenatal screening, yet traditional ultrasound examinations have issues such as reliance on the operator’s experience and the subjectivity of image interpretation. The integration of AI has led to significant advancements in ultrasound examination. For instance, in cleft lip detection, studies have shown that AI algorithms incorporating spatiotemporal facial features can achieve an accuracy of 88%, an improvement over models relying solely on static features (84%).21 22 Furthermore, AI technology can automatically measure fetal biometric parameters and analyse them to more accurately estimate fetal weight and gestational age, and promptly identify issues such as fetal growth retardation or excessive growth.23 24 AI can also assist in identifying the structure of each fetal organ, aiding doctors in more accurately detecting whether fetuses have organ malformations or developmental abnormalities.25
Electronic fetal monitoring
Electronic fetal monitoring is an essential tool for assessing the condition of fetuses in utero; however, traditional methods depend heavily on the expertise of doctors and are susceptible to human error. AI technology offers a more objective and precise approach to fetal monitoring. For instance, a hybrid model that combines one-dimensional convolutional neural network (1D-CNN) and bidirectional gated recurrent unit (GRU) can process fetal monitoring data in real-time with an accuracy exceeding 95%.26 AI technology can promptly identify potential abnormal conditions, such as fetal hypoxia, and significantly enhance the early warning capability for fetal distress in utero. Furthermore, AI technology can predict the risk of fetal acidosis by conducting in-depth analyses of trends and characteristics in fetal heart rate changes, offering an accurate basis for clinical decision-making by doctors.27
Obstetric surgical assistance
In the field of obstetric surgery, including caesarean sections, AI technology is increasingly playing a significant role. It can perform precise three-dimensional reconstructions and surgical planning based on patients’ imaging data, offering doctors intuitive and accurate surgical plans.28 In robotic surgery, AI uses computer vision technology to accurately identify anatomical landmarks in real-time, thereby reducing the risk of surgical injury.29 Additionally, AI technology can analyse images and data in real-time during surgery, offering real-time surgical guidance and suggestions to doctors, thereby enhancing the accuracy and safety of the procedure.30
