Artificial Intelligence in Feto-maternal Health: A Systematic Review of Predictive Models, Validation, and Clinical Translation

Artificial Intelligence in Feto-maternal Health: A Systematic Review of Predictive Models, Validation, and Clinical Translation | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Artificial Intelligence in Feto-maternal Health: A Systematic Review of Predictive Models, Validation, and Clinical Translation Oluwafunmilola Deborah Awe, Ana Paula Morais e Oliveira, Charles M'poca Charles, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8651301/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 10 You are reading this latest preprint version Abstract Maternal mortality remains a major global challenge, especially in developing countries. This review assessed Artificial Intelligence applications in Feto-Maternal health, focusing on validation, performance, and implementation. Following PRISMA guidelines and PROSPERO registration (CRD42023347209), we searched PubMed, EMBASE, Cochrane Library, and Web of Science for studies (2000–2023) applying AI to maternal or neonatal outcomes with defined methodology and validation. Of 14,049 studies, 72 met inclusion criteria; 38 (52.8%) reported external validation, and 11 (15.3%) involved experimental or interventional use. Most datasets (81.3%) were from high-income countries, mainly Asia and North America. Frequently assessed outcomes included preterm birth, low birth weight, and neonatal mortality. Random Forest and XGBoost were most used. Internal performance was strong (AUC 0.82; accuracy 89%), with modest declines in external validation (AUC 0.80; accuracy 87%) and reduced sensitivity in real-world settings. AI shows promise but requires rigorously validated, context-specific models, ethical oversight, and readiness for LMIC integration. Health sciences/Diseases Health sciences/Health care Health sciences/Medical research Artificial intelligence Maternal health Neonatal outcomes Machine learning Predictive modeling Systematic review Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Introduction Worldwide, a maternal death occurs approximately every two minutes due to pregnancy related complications, most of which are preventable with timely and adequate healthcare interventions [1]. For every maternal death, 20 women experience severe morbidities, and a woman or newborn dies every 11 seconds. These deaths often arise from preventable causes during the pre, peri, and post-natal periods. Maternal and fetal outcomes are deeply interconnected; improving maternal health directly benefits neonatal health. Shockingly, 94% of maternal deaths occur in low- and middle-income countries, with poverty and lack of skilled healthcare as primary contributors (PAHO, Maternal Health, 2024) [1–3]. Direct causes include postpartum hemorrhage, sepsis, hypertensive disorders of pregnancy (e.g., eclampsia), and unsafe abortions, while indirect causes include comorbidities and injuries [2–4]. Over the years, interventions to improve feto-maternal outcomes have included community based primary healthcare [5], community support systems promoting safe motherhood [6], midwife-led continuity models [7], point-of-care diagnostics [8], mobile health (mHealth) tools [9,10], and more recently, the use of Artificial Intelligence (AI) to enhance timely diagnosis and treatment of life-threatening conditions. This highlights the importance of conducting comprehensive reviews to evaluate the applications and impact of AI in maternal and neonatal health, especially during the perinatal period. Healthcare organizations increasingly leverage big data from Electronic Health Records (EHRs) to identify patterns in maternal and neonatal indicators, often imperceptible to human intelligence due to biases. AI algorithms can enhance diagnostic accuracy, mitigate risks, prevent complications, optimize resource mobilization, and enable novel treatment strategies, leading to significant improvements in feto-maternal outcomes [11–13]. Artificial Intelligence is the capacity of machines to perform tasks that traditionally require human intelligence. Machine learning (ML), a sub-discipline of AI, uses statistical techniques to analyze data and make predictions to support decision-making. ML is categorized into three types: supervised, unsupervised, and reinforcement learning [15-16]. Supervised learning relies on labeled data (inputs and outputs) and is highly effective when labels are accurate, while unsupervised learning works with unlabeled data, requiring less human intervention and uncovering previously unknown predictors [17,18]. Common supervised algorithms include decision trees (fast but overfit prone), artificial neural networks (flexible but slow and sensitive to missing data), probabilistic neural networks (accurate but slow), multilayer perceptrons (handle non-linear data but time-consuming), radial basis function networks (efficient for regular surfaces), random forests (resistant to noise but hard to interpret), deep learning networks (high accuracy but complex), support vector machines (accurate but assume linear separability), and k-nearest neighbors (simple but slow with missing data) [18–23]. Unsupervised techniques include hierarchical clustering (accurate but limited for extreme dataset sizes), k-means (efficient but sensitive to noise), and fuzzy c-means clustering (accurate but slow) [18]. Algorithm choice depends on dataset size, noise, computational efficiency, and handling of missing or correlated data. Despite global efforts, maternal and neonatal morbidity and mortality remain disproportionately high in low- and middle-income countries [24]. While AI shows promise in predictive modeling, evidence on its practical application in improving feto-maternal outcomes remains limited. Many studies highlight computational advantages of ML over traditional statistical methods [25], yet few evaluate AI in real world clinical settings, and some suggest no significant differences in outcomes [26]. Notably, literature on interventional applications of AI to predict feto-maternal outcomes using real-world data is scarce, highlighting the need for rigorous evaluation. This study addresses this gap by synthesizing evidence from studies that applied artificial intelligence to maternal and perinatal data, generating actionable insights for real world clinical settings. Methodology Study registration and Design This systematic review was registered on the International Prospective Register of Systematic Reviews (PROSPERO) on the 10/07/2023 (CRD42023347209) under the title: Evidence for effectiveness of Artificial Intelligence to Improve Feto-Maternal Outcomes-A Systematic Review Protocol. The PRISMA framework was utilized to guide the assessment of study quality and the synthesis of evidence in this systematic review [27]. To ensure a comprehensive review, a wide range of cohort, cross-sectional, and case control studies were retrieved from major medical databases. Eligibility Criteria Inclusion Criteria This systematic review included studies that applied artificial intelligence (AI) to improve feto-maternal outcomes. Only original research articles were considered, ensuring that the review focused on primary studies rather than secondary sources such as reviews, commentaries, or editorials. To maintain methodological rigor, studies were required to clearly outline their AI methodologies and validation strategies, ensuring the reliability and applicability of their findings. Additionally, included studies had to report maternal and/or fetal outcomes to assess the direct impact of AI applications on pregnancy and neonatal health. No language restrictions were applied, allowing for a comprehensive and globally representative synthesis of the available evidence. Exclusion Criteria Studies were excluded if they lacked full text, focused on animals, or did not report relevant feto-maternal outcomes. Reviews, preprints, theses, commentaries, proposed studies, and duplicates were also removed. To ensure quality, studies without external validation, those published before 2000, or retracted papers were excluded. Information Sources The databases searched for relevant materials for this systematic review and meta-analysis included EMBASE (22/08/2023), Scopus (10/08/2023), PubMed (22/08/2023), Compendex (23/08/2023), Cochrane (23/08/2023), and Web of Science (22/08/2023). Search Strategy A systematic search was performed across multiple databases using a broad set of keywords and descriptors. The strategy was structured around four thematic domains: population groups (pregnant women, postpartum women, neonates, preterm infants, maternal health); clinical conditions (prematurity, preterm labor, neonatal complications, postpartum disorders, maternal near miss, hypertensive disorders, diabetes, hemorrhage, sepsis, mental health); applied technologies (AI, machine learning, deep learning, decision support, computer vision, NLP, computational intelligence); and methodological approaches (supervised and transfer learning, knowledge acquisition/representation, model calibration, data mining, predictive analytics). (Full details of the search strategy are provided in Table A in the Supplementary Information). Study Selection and Screening Process The study selection followed PRISMA guidelines through a rigorous multistage process involving five reviewers. Duplicates were removed using EndNote and Rayyan, after which records underwent title, abstract, and full text screening in Rayyan to exclude irrelevant studies [28]. At each stage, two reviewers independently assessed eligibility, resolving disagreements through discussion or with additional reviewer input. Full texts were evaluated against predefined criteria emphasizing relevance, full text availability, external validation, study population, and outcomes of interest. Studies involving animals, conference abstracts, preprints, theses, reviews, retracted papers, publications before 2000, or those without an AI component were excluded. The final eligible studies were included for systematic review and synthesis, with data extraction independently performed by two reviewers using a standardized form. Data Collection and Extraction Data collection and extraction were conducted systematically to ensure consistency, accuracy, and comprehensiveness. Two reviewers independently collected information from each study using a standardized extraction form. Extracted data included general study characteristics (title, publication year, research area, study site, and geographical distribution), study design and methodology (study type, sample size, predictors, AI algorithms, validation techniques, comparators, and statistical methods), and maternal and neonatal outcomes. Performance metrics were recorded for both internal and external validation, including accuracy, sensitivity, specificity, precision, recall, F1-score, and AUC with confidence intervals where available. Model calibration and comparisons were also documented, noting whether calibration was reported and the statistical methods used for evaluation. AUC, along with its standard error and confidence intervals, was used as the primary effect measure, as it was the most commonly reported metric across studies. Where information was missing or unclear, it was documented without imputation. This structured approach ensured a thorough, systematic, and reproducible analysis of the included studies. Risk of bias assessment and Applicability Concern Given that all the included studies focused on artificial intelligence (AI) for predicting feto-maternal outcomes, we applied the PROBAST (Prediction model Risk of Bias Assessment Tool) to assess both risk of bias and applicability. Risk of bias was evaluated across four domains: participants (representativeness of study populations), predictors (clarity, consistency, and availability at the time of prediction), outcomes (definition and reliability of maternal and perinatal outcomes), and analysis (handling of missing data, overfitting, and validation methods). Applicability was assessed in terms of whether study populations reflected intended clinical settings, whether predictors were feasible for routine practice, and whether outcomes were clinically relevant for decision making. This structured evaluation highlighted potential limitations, such as selective populations and inconsistent outcome definitions, while ensuring that the evidence synthesized was methodologically robust and clinically meaningful. The tool proved invaluable in supporting the reliability of the AI models analyzed in the review [29,30]. Data Synthesis and Analysis A narrative synthesis was performed to summarize qualitative findings, with studies organized according to methodological characteristics, AI approaches, and maternal versus fetal outcomes. Plots, graphs, and tables were employed to visually and tabularly represent study features, outcome distributions, and trends across the included studies. We also extracted metrics from internal and external validation to assess predictive performance. Internal validation, using techniques such as data splitting, k-fold cross-validation, evaluated model performance within the development dataset and helped minimize overfitting. External validation, by contrast, assessed generalizability in independent datasets, providing insight into reliability and applicability across different populations and real-world settings [31]. Meta-analysis was conducted to combine quantitative results from studies assessing similar AI based prediction models, providing overall estimates of AI’s impact on feto-maternal outcomes. Pooling data across studies generates more precise effect sizes and reduced study variability. To evaluate consistency, we used the I² statistic, which measures heterogeneity across study results. High I² values indicated substantial heterogeneity and the need for caution in interpreting pooled findings, whereas low values reflected greater consistency and reliability of the estimates [32]. The formula for I² is High heterogeneity in predictive model performance is expected because the included studies varied in their populations, study designs, predictors, modeling approaches, validation methods, and outcome definitions. These differences can influence estimates and contribute to variability beyond chance, limiting comparability across studies. For this reason, formal sensitivity analyses and overall certainty assessments (e.g., GRADE) were not conducted; instead, heterogeneity and study specific differences were carefully considered when interpreting both the meta-analytic and narrative results. Software and Computational Environment All analyses were performed exclusively using R (version 4.2). Results The study selection process is visually summarized in the PRISMA flow diagram (Fig. 1 ). 14,049 articles were retrieved from 6 databases. The majority of the studies were sourced from Web of Science (4,699(33.4%)), followed by Embase (3,250 (23.1%)) and Scopus (2,476 (17.6%)), collectively accounting for more than 74% of the total retrieved records. Compendex (1,863 (13.3%)) and PubMed (1,568 (11.2%)) contributed moderately to the total, while Cochrane (193 (1.4%)) provided the fewest records. During the deduplication process, 4,991 duplicate records (35.5%) were removed (1,522 in EndNote and 3,469 in Rayyan), leaving 9,058 records (64.5%) for screening. After deduplication, the articles were screened by title using Rayyan, resulting in the selection of 806 studies for retrieval and review. After screening studies using titles and abstracts, we narrowed the selection to 211.We then assessed full texts for eligibility. After full-text screening, 139 articles (65.9%) were excluded for the following reasons: animal studies (1(0.5%)), conference abstracts (21(10.0%)), lack of external validation (82(38.9%)), article unavailability or incorrect title (8(3.8%)), full paper not accessible (3(1.4%)), incorrect outcome, cohort, or AI focus (10(4.7%)), preprints (3(1.4%)), proposed work or early-stage studies (2(0.9%)), theses (2(0.9%)), retracted papers (1, (0.5%)), studies published before 2000 (4(1.9%)), reviews (1(0.5%)), and duplication (1(0.5%)). Finally, 72 studies were included [ 33 – 104 ], which is 34.1% of the total (Reports assessed for eligibility using Title and Abstract), representing 0.79% of the originally selected 9,058 articles. Of these, 37 studies (51.4%) conducted external validation, representing only 0.79% of the total screened articles after deletion of duplicates (9,058) [ 33 – 70 ]. Additionally, 11(15.3%) implemented AI applications in maternal health through experimental designs, accounting for 0.12% of the total screened studies [ 46 , 62 , 71 – 79 ]. (Further details are provided in Table B in the Supplementary section). Population Size Regarding the population of participants in this study, of the 6,720,839 participants, the largest proportion of data was from North America (81.3%), with a total sample size of 5,469,850. Asia follows with 5.3%, corresponding to 353,515 cases. Europe represents 12.4% of the total, with 834,782 samples. Other continents, such as Africa, South America, Australia, and Intercontinental, contribute much smaller percentages, ranging from 0.03% (Africa) to 0.50% (Not Specified), with sample sizes ranging from 28 to 33,660. These figures provide an overview of the distribution of the dataset across continents, highlighting the disproportionate representation from North America (Table 1 ). Table 1 Distribution of Studies by Continent (Frequency and Percentage) Continent N (%) Asia 353,515(5.3) N. America 5,469,850(81.3) Europe 834,782(12.4) Africa 2,152(0.03) S. America 28(0.00) Australia 110(0.00) Intercontinental 26,742(0.40) Not Specified 33,660(0.50) Year of publication Notably, the majority of the studies included in the analysis were published in 2020 , reflecting a significant increase in research activity during that year (Fig. 4 ). Characteristics and Methodologies The reviewed studies primarily focused on neonatal and perinatal outcomes, with many also addressing obstetric and maternal morbidity, including gestational diabetes and related metabolic disorders. Geographically, most research originated from Asia, Europe, and North America, with limited contributions from Africa and South America, highlighting global disparities. Observational designs predominated, while experimental and qualitative studies were less common. Machine learning was the most widely applied AI methodology, followed by clinical decision support systems, biomedical imaging and signal processing, and deep learning models. These findings underscore the need for broader geographic representation and diverse study designs to maximize AI’s impact on maternal health research. Clinical Focus Areas The studies covered a broad range of maternal and perinatal health topics. The primary research areas included Gestational Diabetes and Metabolic Disorders (GDMD), Hypertensive Disorders and Cardiovascular Complications (HDCC), Obstetric Complications and Maternal Morbidity (OCMM), Neonatal and Perinatal Outcomes (NPO), and Postpartum Depression & Mental Health (PDMH). The majority of studies focused on neonatal and perinatal outcomes (NPO) 28(38.90%), followed by obstetric and maternal morbidity 20(27.80%) and gestational diabetes & metabolic disorders 14(19.40%). Fewer studies addressed hypertension & cardiovascular conditions 7(9.72%) and postpartum mental health 3(4.17%), (Fig. 5 ). *Gestational Diabetes and Metabolic Disorders (GDMD), Hypertensive Disorders and Cardiovascular Complications (HDCC), Obstetric Complications and Maternal Morbidity (OCMM) and Postpartum Depression & Mental Health (PDMH). Clinical decision support systems (CDSS) and biomedical signal and image analysis (BSIA), deep learning (DL) Geographical Distribution The study site variable was categorized by continent to provide insights into regional research trends. Most studies originated from Asia 21(29.17%), Europe 20(27.78%), and North America 17(23.61%), while Africa and Australia each 2(2.78%) and South America 1(1.39%) were minimally represented. Study Design Studies were categorized as observational, experimental, qualitative, and only 1 unspecified. Observational studies were the most prevalent, accounting for 55(70.5%), followed by experimental studies with 11(14.1%) while qualitative studies were the least common with 5(6.41%). A study utilized both qualitative and experimental designs 1(1.39%) [ 98 ], while another combined qualitative and observational approaches 1(1.39%) [ 102 ], reflecting a predominant reliance on retrospective and prospective cohort designs to train and validate AI models. Quality Assessment of Included Studies The risk of bias (RoB) assessment (Fig. 2 ) showed that 57 studies (79.20%) were classified as having a low risk of bias, indicating generally strong methodological rigor. A total of 14 studies (19.44%) were rated as having a high risk of bias, reflecting potential weaknesses in study design or reporting that may affect the reliability of their findings. Only a study (1.39%) had an unclear risk of bias, typically due to insufficient or ambiguous reporting. Notably, studies with lower RoB tended to have larger sample sizes, suggesting that better-powered studies may be more likely to adhere to rigorous methodological standards. Regarding applicability (Fig. 3 ), 66 studies (91.70%) were rated as having low concern, demonstrating that their populations, settings, and interventions were well aligned with the review objectives. 5 studies (6.94%) had high applicability concern, and a study (1.39%) were considered unclear due to limited information. The slightly higher proportion of studies with low applicability concern compared to low RoB suggests that even studies with minor methodological limitations were still relevant to the clinical context of the review. Overall, these findings support both the quality and relevance of the included evidence base (Some details of the RoB and applicability are provided in Table C in the Supplementary section). AI Model Applications and Performance AI Model Categories The AI algorithms used in the included studies were categorized into five main groups. The first category consists of machine learning models, such as Random Forest, XGBoost, Support Vector Machines, and Elastic Net. The second category includes deep learning and AI techniques, with notable examples like Convolutional Neural Networks and Recurrent Neural Networks. The third category covers clinical decision support systems, which feature AI-driven clinical guidelines and automated triage systems. The fourth category focuses on natural language processing and AI assistants, incorporating text mining approaches and AI chatbots aimed at improving maternal health. Lastly, the fifth category includes biomedical signal and imaging analysis, with AI-assisted technologies such as ultrasound and fetal monitoring systems. Machine learning models, particularly Random Forest and XGBoost, were the most common 35(44.9%). Clinical decision support systems and biomedical imaging & signal processing were each used in 12(15.4%) studies. Deep learning, including Convolutional and Recurrent Neural Networks, was applied in 10(12.8%) studies. Natural language processing and AI assistants were the least utilized, appearing in 3 (3.84%) studies. Deep learning models, though less frequent, showed strong performance in image-based and sequential data analysis (Table C in the supplementary section). Predictive Performance Metrics Performance evaluation was based on internal and external validation metrics. Internal validation metrics included mean squared error, R-squared, mean absolute error, accuracy, sensitivity, specificity, positive predictive value, negative predictive value, precision, recall, F1-score, area under the curve (AUC), and odds ratio. Confidence intervals were recorded where available. For externally validated studies, similar performance metrics were documented, with additional comparisons drawn between AI-based approaches and conventional methods. (Full details of the findings are provided in Table D in the Supplementary section). Model Calibration and Comparative Performance Few studies reported model calibration, an essential step for ensuring model reliability in clinical applications. Comparative analysis between AI-driven models and traditional statistical methods indicated that AI models generally outperformed conventional approaches in predictive accuracy and diagnostic utility. However, heterogeneity in study design, dataset quality, and validation approaches posed challenges in drawing definitive conclusions about the superiority of AI-based methodologies. *The x-axis represents different model performance metrics- Accuracy, Precision, and Area Under the Curve (AUC), evaluated during internal and external validation phases. The predictive model demonstrated strong performance in internal validation, with high accuracy (0.89), precision (0.81), and recall (0.89). The calibration score was 1.54, indicating perfect agreement between predicted and observed outcomes. The mean squared error (MSE) was low (0.02), and the R² value of 0.65 suggested a good fit. Sensitivity and specificity were 0.79 and 0.89, respectively, while the area under the curve (AUC) was 0.82, reflecting strong discriminatory power. In external validation, performance remained acceptable but showed a decline. Accuracy decreased slightly to 0.87, and calibration was lower (1.28). The MSE increased to 0.04, and the R² dropped to 0.51, suggesting a reduction in predictive strength. Sensitivity was notably lower (0.59), while specificity remained high at 0.92. The AUC also slightly decreased to 0.80, and the positive predictive value (PPV) varied widely, ranging from 0.08 to 1.00, reflecting variability in external datasets. The area under the precision-recall curve (AUPRC) was lower externally (0.40), indicating potential challenges in handling imbalanced data (Fig. 6 ). Overall, the model exhibited strong internal performance but some generalizability limitations, particularly in sensitivity and predictive value when applied to external dataset Forest Plot Test of Heterogeneity and Model Generalizability Internal Validation Across the included studies, the pooled AUC under the common-effect model was 0.85 [0.85, 0.86], indicating strong model discrimination. When applying a random-effects model, the overall pooled AUC was 0.83 [0.79, 0.87], suggesting a slightly wider range of predictive performance across different datasets. The heterogeneity among studies was notably high (I² = 99.6%, p < 0.001), reflecting considerable variability in model performance across different populations and data sources (Fig. 7 ). External Validation For external validation, AI models demonstrated slightly reduced performance compared to internal validation. The pooled AUC under the common-effect model was 0.85 [0.85, 0.86], whereas the random-effects model estimated an overall pooled AUC of 0.83 [0.79, 0.87]. Similar to internal validation, high heterogeneity was observed (I² = 99.7%, p < 0.001) (Table 2 ), suggesting variations in model generalizability across different external datasets (Fig. 8 ). Further details are found in the supplementary section (Figure A-C) Table 2 AUC Model fit, heterogeneity, and pooled effect estimates of the external validation set Model fit Log-likelihood −394·59 Deviance 845·94 AIC 791·19 Corrected AIC (AICc) 791·85 BIC 791·27 Heterogeneity I² , % 99·17 H² 120·85 Q (df = 7) 845·94 p (Q-test) < 0·0001 Effect estimate Pooled effect 0·8780 (0·8752–0·8808) SE 0·0014 Z -value 608·6 p -value < 0·0001 (k = 8) DISCUSSION This systematic review synthesized evidence on the application of artificial intelligence (AI) in maternal health. Of 9,058 articles screened, 72 studies (0.79%) met the inclusion criteria; 37 (51.1%) reported external validation and 11 (15.3%) employed experimental designs. Most studies originated from Asia (29.2%), Europe (27.8%), and North America (23.6%), yet the largest datasets were from North America (81.3%), underscoring regional disparities in data availability. Observational designs predominated (76.4%), with fewer experimental (15.3%) and qualitative (6.9%) studies. Machine learning was the most common AI approach (48.6%), particularly Random Forest and XGBoost. Clinical Decision Support Systems and Biomedical Signal/Image Analysis each accounted for 16.7% of studies, deep learning and other AI systems for 13.9%, and natural language processing for 4.2%. The most frequent outcomes were neonatal and perinatal events (35.9%), maternal morbidity (28.2%), and gestational diabetes (19.2%). Internally validated models showed strong predictive performance (AUC = 0.85), though performance declined with external validation (AUC = 0.81). Substantial heterogeneity (I² > 99%) limited generalizability. Overall, the review highlights the scarcity of externally validated AI models in maternal health, the dominance of high-income country datasets, and the need for methodological standardization to improve reproducibility and clinical impact. This study addresses this gap by synthesizing evidence from studies that applied Artificial Intelligence–driven insights to maternal and perinatal data, providing actionable results for real-world clinical settings. By consolidating diverse AI applications, it highlights how predictive algorithms, when properly validated and ethically deployed, can strengthen early risk detection, optimize decision-making, and ultimately improve maternal and neonatal outcomes. Substantial heterogeneity observed in both internal and external validation indicates that AI models do not perform uniformly across datasets. Variability may arise from differences in populations, dataset sizes, feature selection, and algorithm configurations. While AI models showed strong predictive performance, pooled AUCs > 0.80 in both internal and external validation, performance was slightly lower externally, suggesting recalibration may be needed for new populations. Despite this, AI-based models retained strong discriminatory power, underscoring their clinical potential. To improve generalizability, future work should standardize model development, address dataset variability, and ensure robust external validation. The paucity of externally validated models and experimental AI applications in maternal health, emphasizing the importance of independent validation. Without it, performance may be overestimated due to bias, overfitting, or limited applicability in new settings. External validation ensures reproducibility, detects bias in training data, and evaluates real-world performance, ultimately safeguarding clinical decision-making and improving maternal outcomes [ 105 ]. Machine learning (ML) has transformed healthcare by improving diagnostic accuracy, treatment outcomes, and clinical decision-making. However, compared with traditional statistical methods, ML offers both advantages and challenges. It can process vast, complex datasets, surpassing traditional methods constrained by linear assumptions. Numerous studies demonstrate ML’s superior diagnostic accuracy and efficiency across medical fields [ 56 , 106 – 116 ]. Deep learning excels in diagnostic imaging, particularly ophthalmology and dermatology [ 107 ], and in diabetic retinopathy detection, matching or exceeding human expertise [ 109 ]. ML also shows high precision in cancer metastasis detection [ 108 ]. Additionally, ML can optimize workflows, reduce errors, and speed diagnoses [ 109 ]. Its ability to handle diverse data types—from genomic information to imaging—enhances decision-making. Extreme gradient boosting outperformed logistic regression in predicting postpartum hemorrhage (C-statistic 0.93 vs 0.87) [ 56 ], demonstrating consistent reliability across settings. Despite potential, ML has limitations. In fetal growth studies, ML offered no predictive advantage over logistic regression [ 117 ]. ML relies on large, high-quality datasets, which are often scarce, especially in low-resource settings [ 112 ]. Many models are “black boxes,” limiting interpretability, a critical concern for clinical decision-making [ 112 ]. Clinician resistance persists due to ethical concerns, data security, and fear of job displacement. ML models are not immune to errors; for example, dermatology ML systems misclassified cases due to misleading features [ 113 ], reinforcing that ML should assist, not replace, clinician judgment [ 112 ]. While ML offers strong potential for diagnostics, predictive accuracy, and data processing, challenges such as high data requirements, limited interpretability, and the risk of over-reliance necessitate caution. Traditional statistical methods remain important for their transparency and simplicity. The future likely lies in hybrid approaches that combine ML’s predictive power with classical methods’ reliability and interpretability [ 118 ]. Realizing the full potential of ML in maternal and neonatal care requires not only methodological rigor but also clear regulatory frameworks to ensure safe, transparent, and ethical implementation. Unlike traditional models, AI often lacks standardized validation and approval guidelines, raising concerns about bias, data protection, and ethics. Regulations must ensure efficacy and ethical compliance through collaboration among clinicians, data scientists, policymakers, and regulators. Key considerations include data transparency, model interpretability, and clinical validation. Models must be trained on diverse populations to avoid reinforcing disparities, and interpretability is essential to gain clinician trust. Techniques like SHAP and LIME improve transparency by highlighting feature contributions [ 119 , 120 ]. Clinical validation through trials and external datasets ensures safety and efficacy [ 31 ]. Without clear regulation, unverified AI models may enter clinical workflows, risking patient harm. Frameworks must address bias, fairness, and data privacy [ 121 – 123 ]. One of the major concerns in regulating AI systems is accountability, particularly as machine learning models evolve autonomously, making it difficult to assign responsibility for their decisions. In radiology and other healthcare sectors, AI must adhere to ethical and regulatory standards to ensure fairness, safety, and transparency. A key issue is data privacy, while de-identified data is commonly used for training AI models, there is still a risk of re-identification. Regulatory frameworks must establish clear guidelines on data security to protect patient confidentiality and comply with privacy laws [ 122 , 123 ]. Bias in AI algorithms is another pressing challenge, as training datasets may not adequately represent diverse patient populations, leading to disparities in healthcare outcomes. This is particularly concerning for underserved groups, including those in rural or underfunded settings. Addressing bias requires more inclusive datasets and regulatory frameworks that promote fairness. Organizations such as the International Medical Device Regulators Forum (IMDRF) are working on global guidelines to ensure AI technologies meet high standards of safety, accuracy, and equity [ 123 ]. Beyond improving predictive accuracy, AI is revolutionizing healthcare by enhancing the personalization of treatments. In cardiovascular imaging, for instance, AI streamlines image analysis, segmentation, and automated measurements, improving efficiency and accuracy. However, AI-driven medical devices (MLMDs) raise concerns about transparency, bias, and unforeseen behaviors, necessitating an adaptive regulatory approach. Institutions like the IMDRF advocate for a lifecycle regulatory model, including continuous post-market evaluation to monitor AI safety and efficacy [ 124 ]. Existing regulatory frameworks, such as the Medical Device Regulation (MDR) in Europe and the FDA’s risk-based classification of medical software in the U.S., face challenges in keeping pace with AI advancements. The regulatory landscape remains fragmented, as different countries adopt varying standards, creating difficulties for multinational healthcare providers and AI developers. Global regulatory convergence is essential for streamlining AI implementation, with initiatives like the US-EU Trade and Technology Council’s AI code of conduct representing steps toward harmonization, though progress has been slow [ 121 , 125 ]. Regulatory bodies must evaluate AI not only for predictive accuracy but also for clinical utility, decision-making impact, and patient outcomes. Prospective studies on real-world integration are particularly needed in maternal health. Adoption in resource-limited settings should consider computational demands, healthcare infrastructure, and clinician input to ensure feasibility and practicality. This systematic review demonstrates that AI tools exhibit strong internal predictive performance for key maternal and neonatal outcomes, including preterm birth, low birth weight, and neonatal mortality. However, their real-world effectiveness remains constrained by limited external validation, geographic imbalances in underlying datasets, and modest reductions in model sensitivity when applied in clinical practice. Although only a small proportion of studies reported bedside or interventional implementation, available evidence suggests that, when appropriately integrated, AI can enhance early risk stratification and support timely clinical decision-making. The findings further indicate that optimal AI performance depends on high-quality data, active clinician oversight, and robust regulatory frameworks. Effective clinical translation therefore requires addressing ethical considerations, ensuring model fairness, and developing context-specific tools, particularly in low- and middle-income countries (LMICs), where the burden of maternal and neonatal mortality is greatest. Overall, these results align with the growing body of evidence supporting the potential of AI in predicting feto-maternal outcomes, while also highlighting important methodological limitations. Substantial heterogeneity, inconsistent outcome definitions, and insufficient external validation limited comparability across studies, and the review itself was constrained by the absence of formal certainty grading (e.g., GRADE). Together, these findings underscore the need for standardized methodologies, rigorous validation across diverse populations, and clearer reporting standards to better inform clinical practice and health policy. Conclusion Artificial intelligence and machine learning hold significant promise for improving maternal and neonatal outcomes through enhanced risk prediction and decision support. However, this review underscores that clinical value depends not only on algorithmic accuracy but also on high-quality, representative data, rigorous external validation, and sustained clinician oversight. AI should function as an augmentative tool within integrated care pathways rather than a replacement for clinical judgment. Achieving real-world impact requires addressing regulatory, ethical, and operational challenges, including bias, accountability, data privacy, and contextual adaptability. Coordinated collaboration among clinicians, researchers, developers, and regulators is essential to establish transparent and equitable frameworks for deployment. With standardized methodologies and robust Governance, particularly in low- and middle-income countries where the burden is greatest. AI can progress from experimental models to trustworthy, scalable tools capable of delivering meaningful global health gains. Declarations Grant number Not applicable. Competing Interests The authors declare no competing interests. Funding Statement This research was funded by the Coordination for the Improvement of Higher Education Personnel (CAPES), Brazil, and DeepMind. The funders had no influence on the study design, data collection and analysis, interpretation of results, decision to publish, or preparation of this manuscript. There Author Contribution ODA and RCP conceptualized the study. Methodology was developed by RCP, ODA, and APMO. The search strategy was designed and implemented by APMO, ODA, and RCP. Data curation was carried out by APMO and ODA, with APMO responsible for importing records into Rayyan and, together with ODA, performing deduplication. ODA conducted the data analysis. Validation of the study was undertaken by ODA, CMC, RCP, and KEO. Supervision was provided by RCP and CT. ODA drafted the original manuscript, and all authors (ODA, CMC, APMO, KEO, CT, and RCP) contributed to critical review and editing of the manuscript. All authors approved the final version of the manuscript. Acknowledgement This article is based on research conducted as part of my PhD thesis, which was supported by the Coordination for the Improvement of Higher Education Personnel (CAPES), Brazil, and DeepMind. The funders had no influence on the study design, data collection and analysis, interpretation of results, decision to publish, or preparation of this manuscript. Data Availability Selected data for the included studies are provided as supplementary files. A list of excluded articles after full-text screening is also available in the supplementary material. R codes used for data handling and analysis are available from the authors upon reasonable request. References Onambele, L. et al. Trends, projections, and regional disparities of maternal mortality in Africa (1990–2030): an ARIMA forecasting approach. Epidemiologia 4 , 322–351 (2023). Alamrew, A. et al. Determinants of birth preparedness and complication readiness practice among reproductive-age women in Africa: a systematic review and meta-analysis. BMC Public. Health . 24 , 3154 (2024). Green, G. H. Trends in maternal mortality. N Z. Med. J. 65 , 80–86 (1966). Bhutta, Z. A. & Black, R. E. Global maternal, newborn, and child health—so near and yet so far. N Engl. J. Med. 369 , 2226–2235. https://doi.org/10.1056/NEJMra111185 (2013). Sacks, E. et al. Comprehensive review of the evidence regarding the effectiveness of community-based primary health care in improving maternal, neonatal and child health: 3. neonatal health findings. J. Glob Health . 7 , 010903. https://doi.org/10.7189/jogh.07.010903 (2017). Tobe, R. G., Islam, M. T., Yoshimura, Y. & Hossain Strengthening the community support group to improve maternal and neonatal health seeking behaviors: a cluster-randomized controlled trial in Satkhira District, Bangladesh. PLoS One . 14 , e0212847. https://doi.org/10.1371/journal.pone.0212847 (2019). Mortensen, B. et al. Improving maternal and neonatal health by a midwife-led continuity model of care—an observational study in one governmental hospital in Palestine. EClinicalMedicine 10, 84–91 (2019). https://doi.org/10.1016/j.eclinm.2019.04.003 Majors, C. E., Smith, C. A., Natoli, M. E., Kundrod, K. A. & Richards-Kortum, R. Point-of-care diagnostics to improve maternal and neonatal health in low-resource settings. Lab. Chip . 17 , 3351–3387. https://doi.org/10.1039/C7LC00374A (2017). Tobe, R. G., Haque, S. E., Ikegami, K. & Mori, R. Mobile-health tool to improve maternal and neonatal health care in Bangladesh: a cluster randomized controlled trial. BMC Pregnancy Childbirth . 18 , 102. https://doi.org/10.1186/s12884-018-1714-4 (2018). Ilozumba, O. et al. Targeting strategies of mHealth interventions for maternal health in low and middle-income countries: a systematic review protocol. BMJ Open. 8 , e019345. https://doi.org/10.1136/bmjopen-2017-019345 (2018). Khan, M. S., Bates, D. & Kovacheva, V. P. The quest for equitable health care: the potential for artificial intelligence. NEJM Catalyst Innov. Care Deliv . 2 , 14 (2021). Iftikhar, P., Kuijpers, M. V., Khayyat, A., Iftikhar, A. & De Sa, M. D. Artificial intelligence: a new paradigm in obstetrics and gynecology research and clinical practice. Cureus 12 , e7124. https://doi.org/10.7759/cureus.7124 (2020). Ramakrishnan, R., Rao, S. & He, J. R. Perinatal health predictors using artificial intelligence: a review. Womens Health (Lond Engl) . 17 , 17455065211046132. https://doi.org/10.1177/17455065211046132 (2021). Delanerolle, G. et al. Artificial intelligence: a rapid case for advancement in the personalization of gynaecology/obstetric and mental health care. Womens Health (Lond Engl) . 17 , 17455065211018111. https://doi.org/10.1177/17455065211018111 (2021). Davidson, L. & Boland, M. R. Towards deep phenotyping pregnancy: a systematic review on artificial intelligence and machine learning methods to improve pregnancy outcomes. Brief. Bioinform . 22 , bbaa369. https://doi.org/10.1093/bib/bbaa369 (2021). Panch, T., Szolovits, P. & AR. &. Artificial intelligence, machine learning and health systems. J. Glob Health https://doi.org/10.7189/jogh.08.020303 Obrzut, B., Kusy, M., Semczuk, A., Obrzut, M. & Kluska, J. Prediction of 10-year overall survival in patients with operable cervical cancer using a probabilistic neural network. J. Cancer . 10 , 4189–4195. https://doi.org/10.7150/jca.33945 (2019). Rahmani, A. M. et al. ANR. Machine learning (ML) in medicine: review, applications, and challenges. Mathematics. Rutkowski, L. Springer,. Introduction to probabilistic neural networks. In Probabilistic Neural Networks 21–57 (2004). https://doi.org/10.1007/978-3-540-40046-2_3 Satapathy, S. K., Dehuri, S., Jagadev, A. K. & Mishra, S. EEG brain signal classification for epileptic seizure disorder detection. In EEG Brain Signal Classification for Epileptic Seizure Disorder Detection 1–126 (2019). https://doi.org/10.1016/C2018-0-01888-5 Çolakoğlu, N. & Akkaya, B. Comparison of multi-class classification algorithms on early diagnosis of heart diseases recommendation system for Spotify. In Proceedings of the Conference: Recent Advances in Data Science and Business Analytics 162–171 (2019). Sharkawy, A. N. Principle of neural network and its main types (review). J. Adv. Appl. Comput. Math. 7 , 8–19. https://doi.org/10.15377/2409-5761.2020.07.2 (2020). Wang, X., Ding, Y. & Shao, H. The improved radial basis function neural network and its application. Artif. Life Robot . 2 , 8–11. https://doi.org/10.1007/BF02471145 (1998). Goldenberg, R. L., McClure, E. M. & Saleem, S. Improving pregnancy outcomes in low- and middle-income countries. Reprod. Health . 15 , 88. https://doi.org/10.1186/s12978-018-0524-5 (2018). Baeta, T., Rocha, A. L. L., Oliveira, J. A., Couto da Silva, A. P. & Reis, Z. S. N. Accuracy of machine learning and traditional statistical models in the prediction of postpartum haemorrhage: a systematic review. BMJ Open. 15 , e094455. https://doi.org/10.1136/bmjopen-2024-094455 (2025). Gravesteijn, B. Y. et al. Machine learning algorithms performed no better than regression models for prognostication in traumatic brain injury. J. Clin. Epidemiol. 122 , 95–107. https://doi.org/10.1016/j.jclinepi.2020.03.005 (2020). Moher, D. et al. Preferred reporting items for systematic review and meta-analysis protocols (PRISMA-P) 2015 statement. Syst. Rev. 4 , 1. https://doi.org/10.1186/2046-4053-4-1 (2015). Yu, F., Liu, C. & Sharmin, S. Performance, usability, and user experience of Rayyan for systematic reviews. Proc. Assoc. Inf. Sci. Technol. 59, 843–844 (2022). https://doi.org/10.1002/pra2.745 Moons, K. G. M. et al. PROBAST: a tool to assess risk of bias and applicability of prediction model studies: explanation and elaboration. Ann. Intern. Med. 170 , W1–W33. https://doi.org/10.7326/M18-1377 (2019). Ranjbar, A. et al. Machine learning models for predicting preeclampsia: a systematic review. BMC Pregnancy Childbirth . 24 , 6. https://doi.org/10.1186/s12884-023-06220-1 (2024). Steyerberg, E. W. & Harrell, F. E. Jr. Prediction models need appropriate internal, internal-external, and external validation. J. Clin. Epidemiol. 69 , 245–247. https://doi.org/10.1016/j.jclinepi.2015.04.005 (2016). Xu, H., Platt, R. W., Luo, Z. C., Wei, S. & Fraser, W. D. Exploring heterogeneity in meta-analyses: needs, resources and challenges. Paediatr. Perinat. Epidemiol. 22 , 18–28. https://doi.org/10.1111/j.1365-3016.2007.00908.x (2008). Shen, J. et al. An innovative artificial intelligence-based app for the diagnosis of gestational diabetes mellitus (GDM-AI): development study. J. Med. Internet Res. 22 , e21573. https://doi.org/10.2196/21573 (2020). Sufriyana, H., Wu, Y. W. & Su, E. C. Y. Artificial intelligence-assisted prediction of preeclampsia: development and external validation of a nationwide health insurance dataset of the BPJS Kesehatan in Indonesia. EBioMedicine 54, 102710 (2020). https://doi.org/10.1016/j.ebiom.2020.102710 Tsur, A., Batsry, L., Rosenstein, M. G. & Barak, O. Development and validation of a machine-learning model for prediction of shoulder dystocia. Ultrasound Obstet. Gynecol. 588–596. https://doi.org/10.1002/uog.21878 (2020). Ruppel, H. et al. Development and validation of an obstetric comorbidity risk score for clinical use. Womens Health Rep. (New Rochelle) . 2 , 507–515. https://doi.org/10.1089/whr.2021.0046 (2021). Fotiadou, E., Konopczyński, T. J. H. & RV. End-to-end trained encoder–decoder convolutional neural network for fetal electrocardiogram signal denoising. Physiol. Meas. https://doi.org/10.1088/1361-6579/ab69b9 (2021). Hu, T. et al. Establishment of a model for predicting the outcome of induced labor in full-term pregnancy based on machine learning algorithm. Sci. Rep. 12 , 19063. https://doi.org/10.1038/s41598-022-21954-2 (2022). Clapp, M. A., McCoy, T. H., James, K. E., Perlis, R. H. & Kaimal, A. J. Natural language processing of admission notes to predict postpartum hemorrhage. Am. J. Obstet. Gynecol. 228 , S534–S535. https://doi.org/10.1016/j.ajog.2022.11.913 (2023). Li, Y. X. et al. Novel electronic health records applied for prediction of pre-eclampsia: machine-learning algorithms. Pregnancy Hypertens. 26 , 102–109. https://doi.org/10.1016/j.preghy.2021.10.006 (2021). Artzi, N. S. et al. Prediction of gestational diabetes based on nationwide electronic health records. Nat. Med. https://doi.org/10.1038/s41591-019-0724-8 (2020). Li, Y. X., Liu, Y. C., Wang, M. & Huang, Y. L. Prediction of gestational diabetes mellitus at the first trimester: machine-learning algorithms. Arch. Gynecol. Obstet. 309 , 2557–2566. https://doi.org/10.1007/s00404-023-07131-4 (2024). Shah, S. Y. et al. Prediction of postpartum hemorrhage (PPH) using machine learning algorithms in a Kenyan population. Front. Glob Womens Health . 4 , 1161157. https://doi.org/10.3389/fgwh.2023.1161157 (2023). Lipschuetz, M. et al. Prediction of vaginal birth after cesarean deliveries using machine learning. Am. J. Obstet. Gynecol. 222 , 613e1–613e12. https://doi.org/10.1016/j.ajog.2019.12.267 (2020). Li, S. et al. Improving preeclampsia risk prediction by modeling pregnancy trajectories from routinely collected electronic medical record data. NPJ Digit. Med. 5 , 68. https://doi.org/10.1038/s41746-022-00612-x (2022). Qiu, J. Fetal movement detection by wearable accelerometer duo based on machine learning. IEEE Sens. J. https://doi.org/10.1109/JSEN.2022.3172451 (2022). Douali, N., Dollon, J. & Jaulent, M. C. Personalized prediction of gestational diabetes using a clinical decision support system. In IEEE International Conference on Fuzzy Systems (FUZZ-IEEE) 1–5 (2015). (2015). https://doi.org/10.1109/FUZZ-IEEE.2015.7337813 Gu, X. et al. Machine learning approach for the prediction of macrosomia. Vis. Comput. Ind. Biomed. Art . 7 , 22. https://doi.org/10.1186/s42492-024-00172-9 (2024). Yang, J. et al. Machine learning-based risk stratification for gestational diabetes management. Sensors https://doi.org/10.3390/s22134805 (2022). Starling, S. A molecular signal for preeclampsia. Nat. Rev. Endocrinol. 16 , 471–472 (2020). Ahmed, M., Kashem, M. A., Rahman, M. & Khatun, S. Review and analysis of risk factor of maternal health in remote area using the internet of things (IoT). In Lecture Notes in Computer Science, vol. 632, 1–10 (Springer, Singapore, https://doi.org/10.1007/978-981-15-2317-5_30 (2020). Wang, N. et al. Development and validation of risk prediction models for gestational diabetes mellitus using four different methods. Metabolites 12 , 1040. https://doi.org/10.3390/metabo12111040 (2022). Mazloom, A. R. et al. Noninvasive prenatal detection of sex chromosomal aneuploidies by sequencing circulating cell-free DNA from maternal plasma. Prenat Diagn. 33 , 591–597. https://doi.org/10.1002/pd.4127 (2013). Hoffman, M. K., Ma, N. & Roberts, A. A machine learning algorithm for predicting maternal readmission for hypertensive disorders of pregnancy. Am. J. Obstet. Gynecol. MFM . 3 , 100250. https://doi.org/10.1016/j.ajogmf.2020.100250 (2021). Payne, J. L. et al. DNA methylation biomarkers prospectively predict both antenatal and postpartum depression. Psychiatry Res. 285 , 112711. https://doi.org/10.1016/j.psychres.2019.112711 (2020). Venkatesh, K. K. et al. Machine learning and statistical models to predict postpartum hemorrhage. Obstet. Gynecol. 135 , 935–944. https://doi.org/10.1097/AOG.0000000000003759 (2020). Zhong, Q. Y. et al. Use of natural language processing in electronic medical records to identify pregnant women with suicidal behavior: towards a solution to the complex classification problem. Eur. J. Epidemiol. 34 , 153–162. https://doi.org/10.1007/s10654-018-0470-0 (2019). Badr, D. A., De Lucia, F., Carlin, A., Jani, J. C., Cannie, M. M. & Chest CT scan predictors of intensive care unit admission in hospitalized pregnant women with COVID-19: a case-control study. J. Matern Fetal Neonatal Med. 36 , 2241107. https://doi.org/10.1080/14767058.2023.2241107 (2023). Leonard, S. A. et al. Obstetric comorbidity scores and disparities in severe maternal morbidity across marginalized groups. Am. J. Obstet. Gynecol. MFM . 4 , 100530. https://doi.org/10.1016/j.ajogmf.2021.100530 (2022). Duval, A. et al. A hybrid artificial intelligence model leverages multi-centric clinical data to improve fetal heart rate pregnancy prediction across time-lapse systems. Hum. Reprod. 38 , 596–608. https://doi.org/10.1093/humrep/dead023 (2023). Guedalia, J. et al. Transporting an artificial intelligence model to predict emergency cesarean delivery: overcoming challenges posed by interfacility variation. J. Med. Internet Res. 23 , e28120. https://doi.org/10.2196/28120 (2021). Krupa, J. D., Dhanalakshmi, A., R, K. & S. & An improved parallel sub-filter adaptive noise canceler for the extraction of fetal ECG. Biomed. Tech. (Berl) . 66 , 503–514. https://doi.org/10.1515/bmt-2020-0313 (2021). Mateen, B. A., David, A. L. & Denaxas, S. Electronic health records to predict gestational diabetes risk. Trends Pharmacol. Sci. 41 , 301–304. https://doi.org/10.1016/j.tips.2020.03.003 (2020). Park, T. J. et al. Machine learning model for classifying the results of fetal cardiotocography conducted in high-risk pregnancies. Yonsei Med. J. 63 , 692–700. https://doi.org/10.3349/ymj.2022.63.7.692 (2022). Pinto, Y. et al. Gestational diabetes is driven by microbiota-induced inflammation months before diagnosis. Gut 72 , 918–928. https://doi.org/10.1136/gutjnl-2022-328406 (2023). Qiao, S. et al. FLDS: An intelligent feature learning detection system for visualizing medical images supporting fetal four-chamber views. IEEE J. Biomed. Health Inf. 26 , 4814–4825. https://doi.org/10.1109/JBHI.2021.3091579 (2022). Andreasen, L. A. et al. Multi-centre deep learning for placenta segmentation in obstetric ultrasound with multi-observer and cross-country generalization. Sci. Rep. 13 , 2221. https://doi.org/10.1038/s41598-023-29105-x (2023). Bonet-Carne, E. et al. Quantitative ultrasound texture analysis of fetal lungs to predict neonatal respiratory morbidity. Ultrasound Obstet. Gynecol. 45 , 427–433. https://doi.org/10.1002/uog.13441 (2015). Alzubaidi, M. et al. Ensemble transfer learning for fetal head analysis: from segmentation to gestational age and weight prediction. Diagnostics (Basel) . 12 , 1–28. https://doi.org/10.3390/diagnostics12092229 (2022). Leonard, S. A., Kennedy, C. J., Carmichael, S. L., Lyell, D. J. & Main, E. K. An expanded obstetric comorbidity scoring system for predicting severe maternal morbidity. Obstet. Gynecol. 136 , 440–449. https://doi.org/10.1097/AOG.0000000000004022 (2020). Afrizal, S. H. et al. A user-centered design of natural language processing for maternal monitoring chatbot system. In 2022 International Conference on Informatics, Multimedia, Cyber and Information System (ICIMCIS) 244–248 (2022). https://doi.org/10.1109/ICIMCIS56303.2022.10017517 Caballero-Ruiz, E. et al. A web-based clinical decision support system for gestational diabetes: automatic diet prescription and detection of insulin needs. Int. J. Med. Inf. 102 , 35–49. https://doi.org/10.1016/j.ijmedinf.2017.02.014 (2017). Duysburgh, E. et al. Quality of antenatal and childbirth care in rural health facilities in Burkina Faso, Ghana and Tanzania: an intervention study. Trop. Med. Int. Health . 21 , 70–83. https://doi.org/10.1111/tmi.12627 (2016). Brocklehurst, P. et al. Computerised interpretation of fetal heart rate during labour (INFANT): a randomised controlled trial. Lancet 389 , 1719–1729. https://doi.org/10.1016/S0140-6736(17)30568-8 (2017). McDonald, S. C., Brooker, G., Phipps, H. & Hyett, J. The identification and tracking of uterine contractions using template based cross-correlation. Ann. Biomed. Eng. 45 , 2196–2210. https://doi.org/10.1007/s10439-017-1873-x (2017). Carroll, A. E., Biondich, P., Anand, V., Dugan, T. M. & Downs, S. M. A randomized controlled trial of screening for maternal depression with a clinical decision support system. J. Am. Med. Inf. Assoc. 20 , 311–316. https://doi.org/10.1136/amiajnl-2011-000682 (2013). Alabi, O. et al. Robust fetoscopic mosaicking from deep learned flow fields. Int. J. Comput. Assist. Radiol. Surg. 17 , 1125–1134. https://doi.org/10.1007/s11548-022-02623-1 (2022). Liu, Y. et al. Therapeutic effect of cervical cerclage on cervical insufficiency during pregnancy analyzed by magnetic resonance image under neural network algorithm. Sci. Program. 1–9. https://doi.org/10.1155/2021/5879255 (2021). Marques, J. A. L. et al. IoT-based smart health system for ambulatory maternal and fetal monitoring. IEEE Internet Things J. 8 , 16814–16824. https://doi.org/10.1109/JIOT.2020.3037759 (2021). Hawken, S. et al. External validation of machine learning models including newborn metabolomic markers for postnatal gestational age estimation in East and South-East Asian populations. Gates Open. Res. (2024). Sendra-Balcells, C. et al. Generalisability of fetal ultrasound deep learning models to low-resource imaging settings in five African countries. Sci. Rep. 13 , 1–10. https://doi.org/10.1038/s41598-023-29490-3 (2023). Escobar, G. J. et al. Prediction of obstetrical and fetal complications using automated electronic health record data. Am. J. Obstet. Gynecol. 224 , 137–147e7. https://doi.org/10.1016/j.ajog.2020.10.030 (2021). Neves, M. R. et al. Causal dynamic Bayesian networks for the management of glucose control in gestational diabetes. In. IEEE 9th International Conference on Healthcare Informatics (ICHI) 31–40 (2021). (2021). https://doi.org/10.1109/ICHI52183.2021.00018 Oliver, A. S. & Ashwanthika, U. A. R. Labor prediction in pregnant woman based on electromyogram and electrohysterogram using machine learning techniques. IEEE ICISC (2020). Espinosa, C. A. et al. Multiomic signals associated with maternal epidemiological factors contributing to preterm birth in low- and middle-income countries. Sci. Adv. 9 , eade7692. https://doi.org/10.1126/sciadv.ade7692 (2023). Naimi, A. I., Platt, R. W. & Larkin, J. C. Machine learning for fetal growth prediction. Epidemiology 29 , 290–298. https://doi.org/10.1097/EDE.0000000000000788 (2018). Joglekar, M. V. et al. Postpartum circulating microRNA enhances prediction of future type 2 diabetes in women with previous gestational diabetes. Diabetologia 64 , 1516–1526. https://doi.org/10.1007/s00125-021-05429-z (2021). Edelman, E. A. et al. Implementation of an electronic genomic and family health history tool in primary prenatal care. Am. J. Med. Genet. C Semin Med. Genet. 166C , 34–44. https://doi.org/10.1002/ajmg.c.31389 (2014). Lowery, D. et al. OB HUB: Remote electronic fetal monitoring surveillance. MCN Am. J. Matern Child. Nurs. 48 , 62–68. https://doi.org/10.1097/NMC.0000000000000891 (2023). Luo, D. et al. A prenatal ultrasound scanning approach: one-touch technique in second and third trimesters. Ultrasound Med. Biol. 47 , 2258–2265. https://doi.org/10.1016/j.ultrasmedbio.2021.04.020 (2021). Ma, M. et al. Diagnostic performance of fetal intelligent navigation echocardiography (FINE) in fetuses with double-outlet right ventricle (DORV). Int. J. Cardiovasc. Imaging . 36 , 2165–2172. https://doi.org/10.1007/s10554-020-01932-3 (2020). Ocak, H. A medical decision support system based on support vector machines and the genetic algorithm for the evaluation of fetal well-being. J. Med. Syst. 37 , 9913. https://doi.org/10.1007/s10916-012-9913-4 (2013). Perez-Gonzalez, J., Arámbula Cosío, F. & Huegel, J. C. & Medina-Bañuelos, V. Probabilistic learning coherent point drift for 3D ultrasound fetal head registration. Comput. Math. Methods Med. 4271519 (2020). (2020). https://doi.org/10.1155/2020/4271519 Pluym, I. D. et al. Accuracy of automated three-dimensional ultrasound imaging technique for fetal head biometry. Ultrasound Obstet. Gynecol. 57 , 798–803. https://doi.org/10.1002/uog.22171 (2021). Rigla, M., Martínez-Sarriegui, I., García-Sáez, G., Pons, B. & Hernando, M. E. Gestational diabetes management using smart mobile telemedicine. J. Diabetes Sci. Technol. 12 , 260–264. https://doi.org/10.1177/1932296817704442 (2018). Zheng, X., Li, X., Xu, J. & Wei, Y. Intelligent recognition algorithm-based color Doppler ultrasound in the treatment of dangerous placenta previa. J. Healthc. Eng. 9886521 (2021). (2021). https://doi.org/10.1155/2021/9886521 Fung, R. et al. International Fetal and Newborn Growth Consortium for the 21st Century (INTERGROWTH-21st). Achieving accurate estimates of fetal gestational age and personalised predictions of fetal growth based on data from an international prospective cohort study: a population-based machine learning study. Lancet Digit. Health . 2 , e368–e375. https://doi.org/10.1016/S2589-7500(20)30131-X (2020). Montenegro, J. L. Z., da Costa, C. A. & Janssen, L. P. Evaluating the use of chatbot during pregnancy: a usability study. Healthc. Anal. (N Y) . 2 , 100072. https://doi.org/10.1016/j.health.2022.100072 (2022). Wu, L. et al. Fetal ultrasound image quality assessment with deep convolutional networks. IEEE Trans. Cybern . 47 , 1336–1349. https://doi.org/10.1109/TCYB.2017.2671898 (2017). Carlisle, N. et al. Clinicians’ experiences of using and implementing a medical mobile phone app (QUiPP V2) designed to predict the risk of preterm birth and aid clinical decision making. BMC Med. Inf. Decis. Mak. 21 , 320. https://doi.org/10.1186/s12911-021-01681-w (2021). Chuang, L., Hwang, J. Y., Chang, C. H., Yu, C. H. & Chang, F. M. Ultrasound estimation of fetal weight with the use of computerized artificial neural network model. Ultrasound Med. Biol. 28 , 991–996. https://doi.org/10.1016/S0301-5629(02)00554-9 (2002). Flaucher, M. et al. Smartphone-based colorimetric analysis of urine test strips for at-home prenatal care. IEEE J. Transl Eng. Health Med. 10 , 2800109. https://doi.org/10.1109/JTEHM.2022.3179147 (2022). Hernando, M. E., Gómez, E. J., Corcoy, R. & del Pozo, F. Evaluation of DIABNET, a decision support system for therapy planning in gestational diabetes. Comput. Methods Programs Biomed. 62 , 235–248. https://doi.org/10.1016/S0169-2607(00)00070-5 (2000). Pietrolucci, M. E. et al. Evaluation of an artificial intelligent algorithm (Heartassist™) to automatically assess the quality of second trimester cardiac views: a prospective study. J. Perinat. Med. 51 , 920–924. https://doi.org/10.1515/jpm-2023-0052 (2023). Ramspek, C. L., Jager, K. J., Dekker, F. W., Zoccali, C. & van Diepen, M. External validation of prognostic models: what, why, how, when and where? Clin. Kidney J. 14 , 49–58. https://doi.org/10.1093/ckj/sfaa188 (2020). Machine Learning and the Profession of Medicine. JAMA 316 , 4–5 (2016). Gulshan, V. et al. Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs. JAMA 316 , 2402–2410. https://doi.org/10.1001/jama.2016.17216 (2016). Diagnostic Assessment of Deep Learning Algorithms for Detection of Lymph Node Metastases in Women With Breast Cancer. JAMA 318 , 2199–2210 https://doi.org/10.1001/jama.2017.14585 (2017). Topol, E. J. High-performance medicine: the convergence of human and artificial intelligence. Nat. Med. 25 , 44–56. https://doi.org/10.1038/s41591-018-0300-7 (2019). Steele, A. J., Denaxas, S. C., Shah, A. D., Hemingway, H. & Luscombe, N. M. Machine learning models in electronic health records can outperform conventional survival models for predicting patient mortality in coronary artery disease. PLoS One . 13 , e0202344. https://doi.org/10.1371/journal.pone.0202344 (2018). Rajula, H. S. R., Mauri, M. & Fanos, V. Scale-free networks in metabolomics. Bioinformation 14 , 140–144. https://doi.org/10.6026/97320630014140 (2018). Rajula, H. S. R., Verlato, G., Manchia, M., Antonucci, N. & Fanos, V. Comparison of conventional statistical methods with machine learning in medicine: diagnosis, drug development, and treatment. Med. (Lithuania) . 56 , 455. https://doi.org/10.3390/medicina56090455 (2020). Esteva, A. et al. Dermatologist-level classification of skin cancer with deep neural networks. Nature 542 , 115–118. https://doi.org/10.1038/nature21056 (2017). Barnett-Itzhaki, Z. et al. Machine learning vs. classic statistics for the prediction of IVF outcomes. J. Assist. Reprod. Genet. 37 , 2405–2412. https://doi.org/10.1007/s10815-020-01908-1 (2020). Shin, S. et al. Machine learning vs. conventional statistical models for predicting heart failure readmission and mortality. ESC Heart Fail. 8 , 106–115. https://doi.org/10.1002/ehf2.13073 (2021). Song, X., Arnold, M., Jafna, C. & Kenneth, R. Comparison of machine learning techniques with classical statistical models in predicting health outcomes. (2004). https://doi.org/10.3233/978-1-60750-949-3-736 Kuhle, S. et al. Comparison of logistic regression with machine learning methods for the prediction of fetal growth abnormalities: a retrospective cohort study. BMC Pregnancy Childbirth . 18 , 333. https://doi.org/10.1186/s12884-018-1971-2 (2018). Dong, X. L. & Rekatsinas, T. Data integration and machine learning: a natural synergy. In Proceedings of the ACM SIGMOD International Conference on Management of Data 1645–1650 (2018). https://doi.org/10.1145/3183713.3197387 Parisineni, S. R. A. & Pal, M. Enhancing trust and interpretability of complex machine learning models using local interpretable model agnostic SHAP explanations. Int. J. Data Sci. Anal. 18 , 457–466. https://doi.org/10.1007/s41060-023-00458-w (2023). Kalusivalingam, A. K., Sharma, A., Patel, N. & Singh, V. Leveraging SHAP and LIME for enhanced explainability in AI-driven diagnostic systems. 1–23 (2023). Palaniappan, K., Lin, E. Y. T., Vogel, S. & Lim, J. C. W. Gaps in the global regulatory frameworks for the use of artificial intelligence (AI) in the healthcare services sector and key recommendations. Healthc. (Basel) . 12 , 1730. https://doi.org/10.3390/healthcare12171730 (2024). Harvey, H. B. & Gowda, V. Regulatory issues and challenges to artificial intelligence adoption. Radiol. Clin. North. Am. 59 , 1075–1083. https://doi.org/10.1016/j.rcl.2021.07.007 (2021). Pesapane, F., Volonté, C., Codari, M. & Sardanelli, F. Artificial intelligence as a medical device in radiology: ethical and regulatory issues in Europe and the United States. Insights Imaging . 9 , 745–753. https://doi.org/10.1007/s13244-018-0645-y (2018). Wellnhofer, E. Real-world and regulatory perspectives of artificial intelligence in cardiovascular imaging. Front. Cardiovasc. Med. 9 , 890809. https://doi.org/10.3389/fcvm.2022.890809 (2022). Yaeger, K. A., Martini, M., Yaniv, G., Oermann, E. K. & Costa, A. B. United States regulatory approval of medical devices and software applications enhanced by artificial intelligence. Health Policy Technol. 8 , 192–197. https://doi.org/10.1016/j.hlpt.2019.05.006 (2019). Additional Declarations No competing interests reported. Supplementary Files SUPPLEMENTARYFILES1.docx Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 30 Mar, 2026 Reviews received at journal 22 Mar, 2026 Reviews received at journal 17 Mar, 2026 Reviewers agreed at journal 17 Mar, 2026 Reviewers agreed at journal 16 Mar, 2026 Reviewers invited by journal 16 Mar, 2026 Editor invited by journal 22 Jan, 2026 Editor assigned by journal 21 Jan, 2026 Submission checks completed at journal 21 Jan, 2026 First submitted to journal 20 Jan, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8651301","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":607940063,"identity":"2effd620-7408-424e-9967-eef7ee3c5a8f","order_by":0,"name":"Oluwafunmilola Deborah Awe","email":"","orcid":"","institution":"State University of Campinas","correspondingAuthor":false,"prefix":"","firstName":"Oluwafunmilola","middleName":"Deborah","lastName":"Awe","suffix":""},{"id":607940064,"identity":"2ee74207-4058-4bcc-97b6-785db33df68b","order_by":1,"name":"Ana Paula Morais e Oliveira","email":"","orcid":"","institution":"State University of Campinas","correspondingAuthor":false,"prefix":"","firstName":"Ana","middleName":"Paula Morais e","lastName":"Oliveira","suffix":""},{"id":607940065,"identity":"d501e5d2-d002-452e-974b-27985811a561","order_by":2,"name":"Charles M'poca Charles","email":"","orcid":"","institution":"State University of Campinas","correspondingAuthor":false,"prefix":"","firstName":"Charles","middleName":"M'poca","lastName":"Charles","suffix":""},{"id":607940066,"identity":"1b06257a-5333-468c-ac6d-032e807b0174","order_by":3,"name":"Kelechi Elizabeth Oladimeji","email":"","orcid":"","institution":"University of the Witwatersrand","correspondingAuthor":false,"prefix":"","firstName":"Kelechi","middleName":"Elizabeth","lastName":"Oladimeji","suffix":""},{"id":607940067,"identity":"4cbc22c4-d584-4a79-abeb-fd7c53ac986e","order_by":4,"name":"Cristiano Torezzan","email":"","orcid":"","institution":"State University of Campinas","correspondingAuthor":false,"prefix":"","firstName":"Cristiano","middleName":"","lastName":"Torezzan","suffix":""},{"id":607940068,"identity":"d037042f-2c68-4020-b5b9-237af47ec813","order_by":5,"name":"Rodolfo Carvalho Pacagnella","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAyklEQVRIiWNgGAWjYHACA4YHNgwM/EDWATBfghgtCWkMDJINJGsxOADjE9Ji3t688UFCgk2e8Y3cgwd+1GzLY5DufYBXi8yZY8UGCQlpxWY38hIO9hy7Xcwgc9wArxYJiRwzicQfhxO33cgxOMzAdjuxQSINv8PAWhISDidungHS8o8ULRskgFoY24jRwgP1i8SZNwYHe/tuF7PJHCOghR0YYh+AIcbfnmP84ce323n80m34tcBAApzBRpwGZC3E6hgFo2AUjIKRAwCVt0oZl6fj3QAAAABJRU5ErkJggg==","orcid":"","institution":"State University of Campinas","correspondingAuthor":true,"prefix":"","firstName":"Rodolfo","middleName":"Carvalho","lastName":"Pacagnella","suffix":""}],"badges":[],"createdAt":"2026-01-20 16:44:17","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8651301/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8651301/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":105035575,"identity":"08d1eb82-c984-4d74-8c1d-829fc2bab16f","added_by":"auto","created_at":"2026-03-20 07:26:17","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":42888,"visible":true,"origin":"","legend":"\u003cp\u003ePRISMA flow diagram for systematic review\u003c/p\u003e","description":"","filename":"image2.png","url":"https://assets-eu.researchsquare.com/files/rs-8651301/v1/83f7a03458893063b1734977.png"},{"id":105017777,"identity":"ecb810cf-cfa1-4797-9f95-2924a401884e","added_by":"auto","created_at":"2026-03-20 01:21:36","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":10134,"visible":true,"origin":"","legend":"\u003cp\u003eRisk of bias assessment results with majorities of the studies having low risk.\u003c/p\u003e","description":"","filename":"image3.png","url":"https://assets-eu.researchsquare.com/files/rs-8651301/v1/5d5966a4d66cd87b1c215894.png"},{"id":105017775,"identity":"f06905d9-8c21-4da5-b4fc-f0724181a367","added_by":"auto","created_at":"2026-03-20 01:21:36","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":11067,"visible":true,"origin":"","legend":"\u003cp\u003eApplicability Concern bar chart with majorities of the studies having low concern\u003c/p\u003e","description":"","filename":"image4.png","url":"https://assets-eu.researchsquare.com/files/rs-8651301/v1/00d12b6c6ea973b6a0ba90be.png"},{"id":105035817,"identity":"7f6a88b3-4a6f-4188-a148-0e02f474752b","added_by":"auto","created_at":"2026-03-20 07:26:40","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":11855,"visible":true,"origin":"","legend":"\u003cp\u003eLine plot showing the number of publications per year, peaking in 2020.\u003c/p\u003e","description":"","filename":"image5.png","url":"https://assets-eu.researchsquare.com/files/rs-8651301/v1/723dcc2b5dd5ee6ec804729c.png"},{"id":105017781,"identity":"d21f51d6-4831-4fbc-9944-52e5ab6f281a","added_by":"auto","created_at":"2026-03-20 01:21:36","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":61534,"visible":true,"origin":"","legend":"\u003cp\u003eBar chart showing the number of articles by research area (\u003cstrong\u003eneonatal and perinatal outcomes\u003c/strong\u003e (NPO) highest), continent (Asia), study design(observational), and algorithm (Machine Learning (ML))\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e*Gestational Diabetes and Metabolic Disorders (GDMD), Hypertensive Disorders and Cardiovascular Complications (HDCC), Obstetric Complications and Maternal Morbidity (OCMM) and Postpartum Depression \u0026amp; Mental Health (PDMH). Clinical decision support systems (CDSS) and biomedical signal and image analysis (BSIA), deep learning (DL)\u003c/em\u003e\u003c/p\u003e","description":"","filename":"image6.png","url":"https://assets-eu.researchsquare.com/files/rs-8651301/v1/6f1e57fd5d77882d6ddaba07.png"},{"id":105035880,"identity":"13277d4c-a417-4da6-ad97-751893bf27d3","added_by":"auto","created_at":"2026-03-20 07:26:50","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":13375,"visible":true,"origin":"","legend":"\u003cp\u003eBoxplot of the Metrics showing some reduction in performance when externally validated\u003c/p\u003e\n\u003cp\u003e*The x-axis represents different model performance metrics- Accuracy, Precision, and Area Under the Curve (AUC), evaluated during internal and external validation phases.\u003c/p\u003e","description":"","filename":"image7.png","url":"https://assets-eu.researchsquare.com/files/rs-8651301/v1/fa0c28a05e51f22a199bbcad.png"},{"id":105035853,"identity":"78371f24-eedf-4ec0-8273-ee575a8cb455","added_by":"auto","created_at":"2026-03-20 07:26:45","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":43794,"visible":true,"origin":"","legend":"\u003cp\u003eForest Plot and test of heterogeneity for studies that carried out Internal validation with marked heterogeneity\u003c/p\u003e","description":"","filename":"image8.png","url":"https://assets-eu.researchsquare.com/files/rs-8651301/v1/27cd1bfa65c70bdd89ee6753.png"},{"id":105035831,"identity":"e3da02b2-82a8-4992-be6e-dd0c0026fb1f","added_by":"auto","created_at":"2026-03-20 07:26:42","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":36519,"visible":true,"origin":"","legend":"\u003cp\u003eForest Plot and test of heterogeneity for studies that carried out external validation with marked heterogeneity\u003c/p\u003e","description":"","filename":"image9.png","url":"https://assets-eu.researchsquare.com/files/rs-8651301/v1/5c2e2332886aa579ee25c855.png"},{"id":105036869,"identity":"25e447ee-8c96-4cad-a01f-60011c4a389a","added_by":"auto","created_at":"2026-03-20 07:36:28","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1388511,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8651301/v1/28187e35-2b43-4704-af28-523be3e0b126.pdf"},{"id":105017779,"identity":"a90f4c5e-15ef-4197-8921-8da0dfdf8b92","added_by":"auto","created_at":"2026-03-20 01:21:36","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":3536554,"visible":true,"origin":"","legend":"","description":"","filename":"SUPPLEMENTARYFILES1.docx","url":"https://assets-eu.researchsquare.com/files/rs-8651301/v1/bcd52998335349d4eeb73ba6.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"\u003cp\u003eArtificial Intelligence in Feto-maternal Health: A Systematic Review of Predictive Models, Validation, and Clinical Translation\u003c/p\u003e","fulltext":[{"header":"Introduction","content":"\u003cp\u003eWorldwide, a maternal death occurs approximately every two minutes due to pregnancy related complications, most of which are preventable with timely and adequate healthcare interventions [1]. For every maternal death, 20 women experience severe morbidities, and a woman or newborn dies every 11 seconds. These deaths often arise from preventable causes during the pre, peri, and post-natal periods. Maternal and fetal outcomes are deeply interconnected; improving maternal health directly benefits neonatal health.\u003c/p\u003e\n\u003cp\u003eShockingly, 94% of maternal deaths occur in low- and middle-income countries, with poverty and lack of skilled healthcare as primary contributors (PAHO, Maternal Health, 2024)\u0026nbsp;[1–3]. Direct causes include postpartum hemorrhage, sepsis, hypertensive disorders of pregnancy (e.g., eclampsia), and unsafe abortions, while indirect causes include comorbidities and injuries [2–4]. Over the years, interventions to improve feto-maternal outcomes have included community based primary healthcare [5], community support systems promoting safe motherhood [6], midwife-led continuity models [7], point-of-care diagnostics [8], mobile health (mHealth) tools [9,10], and more recently, the use of Artificial Intelligence (AI) to enhance timely diagnosis and treatment of life-threatening conditions.\u003c/p\u003e\n\u003cp\u003eThis highlights the importance of conducting comprehensive reviews to evaluate the applications and impact of AI in maternal and neonatal health, especially during the perinatal period. Healthcare organizations increasingly leverage big data from Electronic Health Records (EHRs) to identify patterns in maternal and neonatal indicators, often imperceptible to human intelligence due to biases. AI algorithms can enhance diagnostic accuracy, mitigate risks, prevent complications, optimize resource mobilization, and enable novel treatment strategies, leading to significant improvements in feto-maternal outcomes [11–13].\u003c/p\u003e\n\u003cp\u003eArtificial Intelligence is the capacity of machines to perform tasks that traditionally require human intelligence. Machine learning (ML), a sub-discipline of AI, uses statistical techniques to analyze data and make predictions to support decision-making. ML is categorized into three types: supervised, unsupervised, and reinforcement learning [15-16]. Supervised learning relies on labeled data (inputs and outputs) and is highly effective when labels are accurate, while unsupervised learning works with unlabeled data, requiring less human intervention and uncovering previously unknown predictors [17,18].\u003c/p\u003e\n\u003cp\u003eCommon supervised algorithms include decision trees (fast but overfit prone), artificial neural networks (flexible but slow and sensitive to missing data), probabilistic neural networks (accurate but slow), multilayer perceptrons (handle non-linear data but time-consuming), radial basis function networks (efficient for regular surfaces), random forests (resistant to noise but hard to interpret), deep learning networks (high accuracy but complex), support vector machines (accurate but assume linear separability), and k-nearest neighbors (simple but slow with missing data) [18–23]. Unsupervised techniques include hierarchical clustering (accurate but limited for extreme dataset sizes), k-means (efficient but sensitive to noise), and fuzzy c-means clustering (accurate but slow) [18]. Algorithm choice depends on dataset size, noise, computational efficiency, and handling of missing or correlated data.\u003c/p\u003e\n\u003cp\u003eDespite global efforts, maternal and neonatal morbidity and mortality remain disproportionately high in low- and middle-income countries [24]. While AI shows promise in predictive modeling, evidence on its practical application in improving feto-maternal outcomes remains limited. Many studies highlight computational advantages of ML over traditional statistical methods [25], yet few evaluate AI in real world clinical settings, and some suggest no significant differences in outcomes [26]. Notably, literature on interventional applications of AI to predict feto-maternal outcomes using real-world data is scarce, highlighting the need for rigorous evaluation.\u003c/p\u003e\n\u003cp\u003eThis study addresses this gap by synthesizing evidence from studies that applied artificial intelligence to maternal and perinatal data, generating actionable insights for real world clinical settings.\u003c/p\u003e"},{"header":"Methodology","content":"\u003cp\u003e\u003cstrong\u003eStudy registration and Design\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis systematic review was registered on the International Prospective Register of Systematic Reviews (PROSPERO) on the\u0026nbsp;10/07/2023\u0026nbsp;(CRD42023347209)\u0026nbsp;under the title: Evidence for effectiveness of Artificial Intelligence to Improve Feto-Maternal Outcomes-A Systematic Review Protocol. The PRISMA framework was utilized to guide the assessment of study quality and the synthesis of evidence in this systematic review [27]. To ensure a comprehensive review, a wide range of cohort, cross-sectional, and case control studies were retrieved from major medical databases.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEligibility Criteria\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eInclusion Criteria\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003c/strong\u003eThis systematic review included studies that applied artificial intelligence (AI) to improve feto-maternal outcomes. Only original research articles were considered, ensuring that the review focused on primary studies rather than secondary sources such as reviews, commentaries, or editorials. To maintain methodological rigor, studies were required to clearly outline their AI methodologies and validation strategies, ensuring the reliability and applicability of their findings. Additionally, included studies had to report maternal and/or fetal outcomes to assess the direct impact of AI applications on pregnancy and neonatal health. No language restrictions were applied, allowing for a comprehensive and globally representative synthesis of the available evidence.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eExclusion Criteria\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eStudies were excluded if they lacked full text, focused on animals, or did not report relevant feto-maternal outcomes. Reviews, preprints, theses, commentaries, proposed studies, and duplicates were also removed. To ensure quality, studies without external validation, those published before 2000, or retracted papers were excluded.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eInformation Sources\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe databases searched for relevant materials for this systematic review and meta-analysis included EMBASE (22/08/2023), Scopus (10/08/2023), PubMed (22/08/2023), Compendex (23/08/2023), Cochrane (23/08/2023), and Web of Science (22/08/2023).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSearch Strategy\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA systematic search was performed across multiple databases using a broad set of keywords and descriptors. The strategy was structured around four thematic domains: population groups (pregnant women, postpartum women, neonates, preterm infants, maternal health); clinical conditions (prematurity, preterm labor, neonatal complications, postpartum disorders, maternal near miss, hypertensive disorders, diabetes, hemorrhage, sepsis, mental health); applied technologies (AI, machine learning, deep learning, decision support, computer vision, NLP, computational intelligence); and methodological approaches (supervised and transfer learning, knowledge acquisition/representation, model calibration, data mining, predictive analytics). (Full details of the search strategy are provided in Table A in the Supplementary Information).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStudy Selection and Screening Process\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study selection followed PRISMA guidelines through a rigorous multistage process involving five reviewers. Duplicates were removed using EndNote and Rayyan, after which records underwent title, abstract, and full text screening in Rayyan to exclude irrelevant studies [28]. At each stage, two reviewers independently assessed eligibility, resolving disagreements through discussion or with additional reviewer input. Full texts were evaluated against predefined criteria emphasizing relevance, full text availability, external validation, study population, and outcomes of interest. Studies involving animals, conference abstracts, preprints, theses, reviews, retracted papers, publications before 2000, or those without an AI component were excluded. The final eligible studies were included for systematic review and synthesis, with data extraction independently performed by two reviewers using a standardized form.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Collection and Extraction\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eData collection and extraction were conducted systematically to ensure consistency, accuracy, and comprehensiveness. Two reviewers independently collected information from each study using a standardized extraction form. Extracted data included general study characteristics (title, publication year, research area, study site, and geographical distribution), study design and methodology (study type, sample size, predictors, AI algorithms, validation techniques, comparators, and statistical methods), and maternal and neonatal outcomes.\u003c/p\u003e\n\u003cp\u003ePerformance metrics were recorded for both internal and external validation, including accuracy, sensitivity, specificity, precision, recall, F1-score, and AUC with confidence intervals where available. Model calibration and comparisons were also documented, noting whether calibration was reported and the statistical methods used for evaluation. AUC, along with its standard error and confidence intervals, was used as the primary effect measure, as it was the most commonly reported metric across studies.\u003c/p\u003e\n\u003cp\u003eWhere information was missing or unclear, it was documented without imputation. This structured approach ensured a thorough, systematic, and reproducible analysis of the included studies.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRisk of bias assessment and Applicability Concern\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eGiven that all the included studies focused on artificial intelligence (AI) for predicting feto-maternal outcomes, we applied the PROBAST (Prediction model Risk of Bias Assessment Tool) to assess both risk of bias and applicability. Risk of bias was evaluated across four domains: participants (representativeness of study populations), predictors (clarity, consistency, and availability at the time of prediction), outcomes (definition and reliability of maternal and perinatal outcomes), and analysis (handling of missing data, overfitting, and validation methods). Applicability was assessed in terms of whether study populations reflected intended clinical settings, whether predictors were feasible for routine practice, and whether outcomes were clinically relevant for decision making. This structured evaluation highlighted potential limitations, such as selective populations and inconsistent outcome definitions, while ensuring that the evidence synthesized was methodologically robust and clinically meaningful. The tool proved invaluable in supporting the reliability of the AI models analyzed in the review [29,30].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Synthesis and Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA narrative synthesis was performed to summarize qualitative findings, with studies organized according to methodological characteristics, AI approaches, and maternal versus fetal outcomes. Plots, graphs, and tables were employed to visually and tabularly represent study features, outcome distributions, and trends across the included studies. We also extracted metrics from internal and external validation to assess predictive performance. Internal validation, using techniques such as data splitting, k-fold cross-validation, evaluated model performance within the development dataset and helped minimize overfitting. External validation, by contrast, assessed generalizability in independent datasets, providing insight into reliability and applicability across different populations and real-world settings [31].\u003c/p\u003e\n\u003cp\u003eMeta-analysis was conducted to combine quantitative results from studies assessing similar AI based prediction models, providing overall estimates of AI\u0026rsquo;s impact on feto-maternal outcomes. Pooling data across studies generates more precise effect sizes and reduced study variability. To evaluate consistency, we used the I\u0026sup2; statistic, which measures heterogeneity across study results. High I\u0026sup2; values indicated substantial heterogeneity and the need for caution in interpreting pooled findings, whereas low values reflected greater consistency and reliability of the estimates [32].\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe formula for I\u0026sup2; is\u003c/p\u003e\n\u003cp\u003e\u003cimg width=\"180\" height=\"50\" src=\"https://myfiles.space/user_files/69519_bce2c0439cd956a6/69519_custom_files/img1773944507.gif\" alt=\"A black and white image of a mathematical equation AI-generated content may be incorrect.\"\u003e\u003c/p\u003e\n\u003cp\u003eHigh heterogeneity in predictive model performance is expected because the included studies varied in their populations, study designs, predictors, modeling approaches, validation methods, and outcome definitions. These differences can influence estimates and contribute to variability beyond chance, limiting comparability across studies. For this reason, formal sensitivity analyses and overall certainty assessments (e.g., GRADE) were not conducted; instead, heterogeneity and study specific differences were carefully considered when interpreting both the meta-analytic and narrative results.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSoftware and Computational Environment\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll analyses were performed exclusively using R (version 4.2).\u0026nbsp;\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eThe study selection process is visually summarized in the PRISMA flow diagram (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). 14,049 articles were retrieved from 6 databases. The majority of the studies were sourced from Web of Science (4,699(33.4%)), followed by Embase (3,250 (23.1%)) and Scopus (2,476 (17.6%)), collectively accounting for more than 74% of the total retrieved records. Compendex (1,863 (13.3%)) and PubMed (1,568 (11.2%)) contributed moderately to the total, while Cochrane (193 (1.4%)) provided the fewest records.\u003c/p\u003e \u003cp\u003eDuring the deduplication process, 4,991 duplicate records (35.5%) were removed (1,522 in EndNote and 3,469 in Rayyan), leaving 9,058 records (64.5%) for screening.\u003c/p\u003e \u003cp\u003eAfter deduplication, the articles were screened by title using Rayyan, resulting in the selection of 806 studies for retrieval and review. After screening studies using titles and abstracts, we narrowed the selection to 211.We then assessed full texts for eligibility. After full-text screening, 139 articles (65.9%) were excluded for the following reasons: animal studies (1(0.5%)), conference abstracts (21(10.0%)), lack of external validation (82(38.9%)), article unavailability or incorrect title (8(3.8%)), full paper not accessible (3(1.4%)), incorrect outcome, cohort, or AI focus (10(4.7%)), preprints (3(1.4%)), proposed work or early-stage studies (2(0.9%)), theses (2(0.9%)), retracted papers (1, (0.5%)), studies published before 2000 (4(1.9%)), reviews (1(0.5%)), and duplication (1(0.5%)).\u003c/p\u003e \u003cp\u003eFinally, 72 studies were included [\u003cspan additionalcitationids=\"CR34 CR35 CR36 CR37 CR38 CR39 CR40 CR41 CR42 CR43 CR44 CR45 CR46 CR47 CR48 CR49 CR50 CR51 CR52 CR53 CR54 CR55 CR56 CR57 CR58 CR59 CR60 CR61 CR62 CR63 CR64 CR65 CR66 CR67 CR68 CR69 CR70 CR71 CR72 CR73 CR74 CR75 CR76 CR77 CR78 CR79 CR80 CR81 CR82 CR83 CR84 CR85 CR86 CR87 CR88 CR89 CR90 CR91 CR92 CR93 CR94 CR95 CR96 CR97 CR98 CR99 CR100 CR101 CR102 CR103\" citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR104\" class=\"CitationRef\"\u003e104\u003c/span\u003e], which is 34.1% of the total (Reports assessed for eligibility using Title and Abstract), representing 0.79% of the originally selected 9,058 articles. Of these, 37 studies (51.4%) conducted external validation, representing only 0.79% of the total screened articles after deletion of duplicates (9,058) [\u003cspan additionalcitationids=\"CR34 CR35 CR36 CR37 CR38 CR39 CR40 CR41 CR42 CR43 CR44 CR45 CR46 CR47 CR48 CR49 CR50 CR51 CR52 CR53 CR54 CR55 CR56 CR57 CR58 CR59 CR60 CR61 CR62 CR63 CR64 CR65 CR66 CR67 CR68 CR69\" citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e70\u003c/span\u003e]. Additionally, 11(15.3%) implemented AI applications in maternal health through experimental designs, accounting for 0.12% of the total screened studies [\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e, \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e62\u003c/span\u003e, \u003cspan additionalcitationids=\"CR72 CR73 CR74 CR75 CR76 CR77 CR78\" citationid=\"CR71\" class=\"CitationRef\"\u003e71\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR79\" class=\"CitationRef\"\u003e79\u003c/span\u003e]. (Further details are provided in Table B in the Supplementary section).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e\n\u003ch3\u003ePopulation Size\u003c/h3\u003e\n\u003cp\u003eRegarding the population of participants in this study, of the 6,720,839 participants, the largest proportion of data was from North America (81.3%), with a total sample size of 5,469,850. Asia follows with 5.3%, corresponding to 353,515 cases. Europe represents 12.4% of the total, with 834,782 samples. Other continents, such as Africa, South America, Australia, and Intercontinental, contribute much smaller percentages, ranging from 0.03% (Africa) to 0.50% (Not Specified), with sample sizes ranging from 28 to 33,660. These figures provide an overview of the distribution of the dataset across continents, highlighting the disproportionate representation from North America (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eDistribution of Studies by Continent (Frequency and Percentage)\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eContinent\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eN (%)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAsia\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e353,515(5.3)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eN. America\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e5,469,850(81.3)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEurope\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e834,782(12.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAfrica\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2,152(0.03)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eS. America\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e28(0.00)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAustralia\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e110(0.00)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIntercontinental\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e26,742(0.40)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNot Specified\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e33,660(0.50)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eYear of publication\u003c/h2\u003e \u003cp\u003eNotably, the majority of the studies included in the analysis were published in \u003cb\u003e2020\u003c/b\u003e, reflecting a significant increase in research activity during that year (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eCharacteristics and Methodologies\u003c/h3\u003e\n\u003cp\u003eThe reviewed studies primarily focused on neonatal and perinatal outcomes, with many also addressing obstetric and maternal morbidity, including gestational diabetes and related metabolic disorders. Geographically, most research originated from Asia, Europe, and North America, with limited contributions from Africa and South America, highlighting global disparities. Observational designs predominated, while experimental and qualitative studies were less common. Machine learning was the most widely applied AI methodology, followed by clinical decision support systems, biomedical imaging and signal processing, and deep learning models. These findings underscore the need for broader geographic representation and diverse study designs to maximize AI\u0026rsquo;s impact on maternal health research.\u003c/p\u003e\n\u003ch3\u003eClinical Focus Areas\u003c/h3\u003e\n\u003cp\u003eThe studies covered a broad range of maternal and perinatal health topics. The primary research areas included Gestational Diabetes and Metabolic Disorders (GDMD), Hypertensive Disorders and Cardiovascular Complications (HDCC), Obstetric Complications and Maternal Morbidity (OCMM), Neonatal and Perinatal Outcomes (NPO), and Postpartum Depression \u0026amp; Mental Health (PDMH).\u003c/p\u003e \u003cp\u003eThe majority of studies focused on neonatal and perinatal outcomes (NPO) 28(38.90%), followed by obstetric and maternal morbidity 20(27.80%) and gestational diabetes \u0026amp; metabolic disorders 14(19.40%). Fewer studies addressed hypertension \u0026amp; cardiovascular conditions 7(9.72%) and postpartum mental health 3(4.17%), (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003e*Gestational Diabetes and Metabolic Disorders (GDMD), Hypertensive Disorders and Cardiovascular Complications (HDCC), Obstetric Complications and Maternal Morbidity (OCMM) and Postpartum Depression \u0026amp; Mental Health (PDMH). Clinical decision support systems (CDSS) and biomedical signal and image analysis (BSIA), deep learning (DL)\u003c/em\u003e \u003c/p\u003e\n\u003ch3\u003eGeographical Distribution\u003c/h3\u003e\n\u003cp\u003eThe study site variable was categorized by continent to provide insights into regional research trends. Most studies originated from Asia 21(29.17%), Europe 20(27.78%), and North America 17(23.61%), while Africa and Australia each 2(2.78%) and South America 1(1.39%) were minimally represented.\u003c/p\u003e\n\u003ch3\u003eStudy Design\u003c/h3\u003e\n\u003cp\u003eStudies were categorized as observational, experimental, qualitative, and only 1 unspecified. Observational studies were the most prevalent, accounting for 55(70.5%), followed by experimental studies with 11(14.1%) while qualitative studies were the least common with 5(6.41%). A study utilized both qualitative and experimental designs 1(1.39%) [\u003cspan citationid=\"CR98\" class=\"CitationRef\"\u003e98\u003c/span\u003e], while another combined qualitative and observational approaches 1(1.39%) [\u003cspan citationid=\"CR102\" class=\"CitationRef\"\u003e102\u003c/span\u003e], reflecting a predominant reliance on retrospective and prospective cohort designs to train and validate AI models.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eQuality Assessment of Included Studies\u003c/h2\u003e \u003cp\u003eThe risk of bias (RoB) assessment (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e) showed that 57 studies (79.20%) were classified as having a low risk of bias, indicating generally strong methodological rigor. A total of 14 studies (19.44%) were rated as having a high risk of bias, reflecting potential weaknesses in study design or reporting that may affect the reliability of their findings. Only a study (1.39%) had an unclear risk of bias, typically due to insufficient or ambiguous reporting. Notably, studies with lower RoB tended to have larger sample sizes, suggesting that better-powered studies may be more likely to adhere to rigorous methodological standards.\u003c/p\u003e \u003cp\u003eRegarding applicability (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e), 66 studies (91.70%) were rated as having low concern, demonstrating that their populations, settings, and interventions were well aligned with the review objectives. 5 studies (6.94%) had high applicability concern, and a study (1.39%) were considered unclear due to limited information. The slightly higher proportion of studies with low applicability concern compared to low RoB suggests that even studies with minor methodological limitations were still relevant to the clinical context of the review. Overall, these findings support both the quality and relevance of the included evidence base (Some details of the RoB and applicability are provided in Table C in the Supplementary section).\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eAI Model Applications and Performance\u003c/h3\u003e\n\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003eAI Model Categories\u003c/h2\u003e \u003cp\u003eThe AI algorithms used in the included studies were categorized into five main groups. The first category consists of machine learning models, such as Random Forest, XGBoost, Support Vector Machines, and Elastic Net. The second category includes deep learning and AI techniques, with notable examples like Convolutional Neural Networks and Recurrent Neural Networks. The third category covers clinical decision support systems, which feature AI-driven clinical guidelines and automated triage systems. The fourth category focuses on natural language processing and AI assistants, incorporating text mining approaches and AI chatbots aimed at improving maternal health. Lastly, the fifth category includes biomedical signal and imaging analysis, with AI-assisted technologies such as ultrasound and fetal monitoring systems.\u003c/p\u003e \u003cp\u003eMachine learning models, particularly Random Forest and XGBoost, were the most common 35(44.9%). Clinical decision support systems and biomedical imaging \u0026amp; signal processing were each used in 12(15.4%) studies. Deep learning, including Convolutional and Recurrent Neural Networks, was applied in 10(12.8%) studies. Natural language processing and AI assistants were the least utilized, appearing in 3 (3.84%) studies. Deep learning models, though less frequent, showed strong performance in image-based and sequential data analysis (Table C in the supplementary section).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003ePredictive Performance Metrics\u003c/h2\u003e \u003cp\u003ePerformance evaluation was based on internal and external validation metrics. Internal validation metrics included mean squared error, R-squared, mean absolute error, accuracy, sensitivity, specificity, positive predictive value, negative predictive value, precision, recall, F1-score, area under the curve (AUC), and odds ratio. Confidence intervals were recorded where available. For externally validated studies, similar performance metrics were documented, with additional comparisons drawn between AI-based approaches and conventional methods. (Full details of the findings are provided in Table D in the Supplementary section).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eModel Calibration and Comparative Performance\u003c/h2\u003e \u003cp\u003eFew studies reported model calibration, an essential step for ensuring model reliability in clinical applications. Comparative analysis between AI-driven models and traditional statistical methods indicated that AI models generally outperformed conventional approaches in predictive accuracy and diagnostic utility. However, heterogeneity in study design, dataset quality, and validation approaches posed challenges in drawing definitive conclusions about the superiority of AI-based methodologies.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e*The x-axis represents different model performance metrics- Accuracy, Precision, and Area Under the Curve (AUC), evaluated during internal and external validation phases.\u003c/p\u003e \u003cp\u003eThe predictive model demonstrated strong performance in internal validation, with high accuracy (0.89), precision (0.81), and recall (0.89). The calibration score was 1.54, indicating perfect agreement between predicted and observed outcomes. The mean squared error (MSE) was low (0.02), and the R\u0026sup2; value of 0.65 suggested a good fit. Sensitivity and specificity were 0.79 and 0.89, respectively, while the area under the curve (AUC) was 0.82, reflecting strong discriminatory power.\u003c/p\u003e \u003cp\u003eIn external validation, performance remained acceptable but showed a decline. Accuracy decreased slightly to 0.87, and calibration was lower (1.28). The MSE increased to 0.04, and the R\u0026sup2; dropped to 0.51, suggesting a reduction in predictive strength. Sensitivity was notably lower (0.59), while specificity remained high at 0.92. The AUC also slightly decreased to 0.80, and the positive predictive value (PPV) varied widely, ranging from 0.08 to 1.00, reflecting variability in external datasets. The area under the precision-recall curve (AUPRC) was lower externally (0.40), indicating potential challenges in handling imbalanced data (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eOverall, the model exhibited strong internal performance but some generalizability limitations, particularly in sensitivity and predictive value when applied to external dataset\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eForest Plot\u003c/h2\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eTest of Heterogeneity and Model Generalizability\u003c/h2\u003e \u003cdiv id=\"Sec15\" class=\"Section3\"\u003e \u003ch2\u003eInternal Validation\u003c/h2\u003e \u003cp\u003eAcross the included studies, the pooled AUC under the common-effect model was 0.85 [0.85, 0.86], indicating strong model discrimination. When applying a random-effects model, the overall pooled AUC was 0.83 [0.79, 0.87], suggesting a slightly wider range of predictive performance across different datasets. The heterogeneity among studies was notably high (I\u0026sup2; = 99.6%, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), reflecting considerable variability in model performance across different populations and data sources (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eExternal Validation\u003c/h2\u003e \u003cp\u003eFor external validation, AI models demonstrated slightly reduced performance compared to internal validation. The pooled AUC under the common-effect model was 0.85 [0.85, 0.86], whereas the random-effects model estimated an overall pooled AUC of 0.83 [0.79, 0.87]. Similar to internal validation, high heterogeneity was observed (I\u0026sup2; = 99.7%, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e), suggesting variations in model generalizability across different external datasets (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e). Further details are found in the supplementary section (Figure A-C)\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eAUC Model fit, heterogeneity, and pooled effect estimates of the external validation set\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel fit\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLog-likelihood\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026minus;394\u0026middot;59\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDeviance\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e845\u0026middot;94\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAIC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e791\u0026middot;19\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCorrected AIC (AICc)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e791\u0026middot;85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBIC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e791\u0026middot;27\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eHeterogeneity\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eI\u0026sup2;\u003c/em\u003e, %\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e99\u0026middot;17\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eH\u0026sup2;\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e120\u0026middot;85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eQ\u003c/em\u003e (df\u0026thinsp;=\u0026thinsp;7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e845\u0026middot;94\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003ep\u003c/em\u003e (Q-test)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0\u0026middot;0001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eEffect estimate\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePooled effect\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u0026middot;8780 (0\u0026middot;8752\u0026ndash;0\u0026middot;8808)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSE\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u0026middot;0014\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eZ\u003c/em\u003e-value\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e608\u0026middot;6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003ep\u003c/em\u003e-value\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0\u0026middot;0001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"3\"\u003e(k\u0026thinsp;=\u0026thinsp;8)\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003eThis systematic review synthesized evidence on the application of artificial intelligence (AI) in maternal health. Of 9,058 articles screened, 72 studies (0.79%) met the inclusion criteria; 37 (51.1%) reported external validation and 11 (15.3%) employed experimental designs. Most studies originated from Asia (29.2%), Europe (27.8%), and North America (23.6%), yet the largest datasets were from North America (81.3%), underscoring regional disparities in data availability.\u003c/p\u003e \u003cp\u003eObservational designs predominated (76.4%), with fewer experimental (15.3%) and qualitative (6.9%) studies. Machine learning was the most common AI approach (48.6%), particularly Random Forest and XGBoost. Clinical Decision Support Systems and Biomedical Signal/Image Analysis each accounted for 16.7% of studies, deep learning and other AI systems for 13.9%, and natural language processing for 4.2%.\u003c/p\u003e \u003cp\u003eThe most frequent outcomes were neonatal and perinatal events (35.9%), maternal morbidity (28.2%), and gestational diabetes (19.2%). Internally validated models showed strong predictive performance (AUC\u0026thinsp;=\u0026thinsp;0.85), though performance declined with external validation (AUC\u0026thinsp;=\u0026thinsp;0.81). Substantial heterogeneity (I\u0026sup2; \u0026gt; 99%) limited generalizability.\u003c/p\u003e \u003cp\u003eOverall, the review highlights the scarcity of externally validated AI models in maternal health, the dominance of high-income country datasets, and the need for methodological standardization to improve reproducibility and clinical impact.\u003c/p\u003e \u003cp\u003eThis study addresses this gap by synthesizing evidence from studies that applied Artificial Intelligence\u0026ndash;driven insights to maternal and perinatal data, providing actionable results for real-world clinical settings. By consolidating diverse AI applications, it highlights how predictive algorithms, when properly validated and ethically deployed, can strengthen early risk detection, optimize decision-making, and ultimately improve maternal and neonatal outcomes.\u003c/p\u003e \u003cp\u003eSubstantial heterogeneity observed in both internal and external validation indicates that AI models do not perform uniformly across datasets. Variability may arise from differences in populations, dataset sizes, feature selection, and algorithm configurations. While AI models showed strong predictive performance, pooled AUCs\u0026thinsp;\u0026gt;\u0026thinsp;0.80 in both internal and external validation, performance was slightly lower externally, suggesting recalibration may be needed for new populations. Despite this, AI-based models retained strong discriminatory power, underscoring their clinical potential. To improve generalizability, future work should standardize model development, address dataset variability, and ensure robust external validation.\u003c/p\u003e \u003cp\u003eThe paucity of externally validated models and experimental AI applications in maternal health, emphasizing the importance of independent validation. Without it, performance may be overestimated due to bias, overfitting, or limited applicability in new settings. External validation ensures reproducibility, detects bias in training data, and evaluates real-world performance, ultimately safeguarding clinical decision-making and improving maternal outcomes [\u003cspan citationid=\"CR105\" class=\"CitationRef\"\u003e105\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eMachine learning (ML) has transformed healthcare by improving diagnostic accuracy, treatment outcomes, and clinical decision-making. However, compared with traditional statistical methods, ML offers both advantages and challenges. It can process vast, complex datasets, surpassing traditional methods constrained by linear assumptions. Numerous studies demonstrate ML\u0026rsquo;s superior diagnostic accuracy and efficiency across medical fields [\u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e56\u003c/span\u003e, \u003cspan additionalcitationids=\"CR107 CR108 CR109 CR110 CR111 CR112 CR113 CR114 CR115\" citationid=\"CR106\" class=\"CitationRef\"\u003e106\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR116\" class=\"CitationRef\"\u003e116\u003c/span\u003e]. Deep learning excels in diagnostic imaging, particularly ophthalmology and dermatology [\u003cspan citationid=\"CR107\" class=\"CitationRef\"\u003e107\u003c/span\u003e], and in diabetic retinopathy detection, matching or exceeding human expertise [\u003cspan citationid=\"CR109\" class=\"CitationRef\"\u003e109\u003c/span\u003e]. ML also shows high precision in cancer metastasis detection [\u003cspan citationid=\"CR108\" class=\"CitationRef\"\u003e108\u003c/span\u003e]. Additionally, ML can optimize workflows, reduce errors, and speed diagnoses [\u003cspan citationid=\"CR109\" class=\"CitationRef\"\u003e109\u003c/span\u003e]. Its ability to handle diverse data types\u0026mdash;from genomic information to imaging\u0026mdash;enhances decision-making. Extreme gradient boosting outperformed logistic regression in predicting postpartum hemorrhage (C-statistic 0.93 vs 0.87) [\u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e56\u003c/span\u003e], demonstrating consistent reliability across settings.\u003c/p\u003e \u003cp\u003eDespite potential, ML has limitations. In fetal growth studies, ML offered no predictive advantage over logistic regression [\u003cspan citationid=\"CR117\" class=\"CitationRef\"\u003e117\u003c/span\u003e]. ML relies on large, high-quality datasets, which are often scarce, especially in low-resource settings [\u003cspan citationid=\"CR112\" class=\"CitationRef\"\u003e112\u003c/span\u003e]. Many models are \u0026ldquo;black boxes,\u0026rdquo; limiting interpretability, a critical concern for clinical decision-making [\u003cspan citationid=\"CR112\" class=\"CitationRef\"\u003e112\u003c/span\u003e]. Clinician resistance persists due to ethical concerns, data security, and fear of job displacement. ML models are not immune to errors; for example, dermatology ML systems misclassified cases due to misleading features [\u003cspan citationid=\"CR113\" class=\"CitationRef\"\u003e113\u003c/span\u003e], reinforcing that ML should assist, not replace, clinician judgment [\u003cspan citationid=\"CR112\" class=\"CitationRef\"\u003e112\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eWhile ML offers strong potential for diagnostics, predictive accuracy, and data processing, challenges such as high data requirements, limited interpretability, and the risk of over-reliance necessitate caution. Traditional statistical methods remain important for their transparency and simplicity. The future likely lies in hybrid approaches that combine ML\u0026rsquo;s predictive power with classical methods\u0026rsquo; reliability and interpretability [\u003cspan citationid=\"CR118\" class=\"CitationRef\"\u003e118\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eRealizing the full potential of ML in maternal and neonatal care requires not only methodological rigor but also clear regulatory frameworks to ensure safe, transparent, and ethical implementation. Unlike traditional models, AI often lacks standardized validation and approval guidelines, raising concerns about bias, data protection, and ethics.\u003c/p\u003e \u003cp\u003eRegulations must ensure efficacy and ethical compliance through collaboration among clinicians, data scientists, policymakers, and regulators. Key considerations include data transparency, model interpretability, and clinical validation. Models must be trained on diverse populations to avoid reinforcing disparities, and interpretability is essential to gain clinician trust. Techniques like SHAP and LIME improve transparency by highlighting feature contributions [\u003cspan citationid=\"CR119\" class=\"CitationRef\"\u003e119\u003c/span\u003e, \u003cspan citationid=\"CR120\" class=\"CitationRef\"\u003e120\u003c/span\u003e]. Clinical validation through trials and external datasets ensures safety and efficacy [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. Without clear regulation, unverified AI models may enter clinical workflows, risking patient harm. Frameworks must address bias, fairness, and data privacy [\u003cspan additionalcitationids=\"CR122\" citationid=\"CR121\" class=\"CitationRef\"\u003e121\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR123\" class=\"CitationRef\"\u003e123\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eOne of the major concerns in regulating AI systems is accountability, particularly as machine learning models evolve autonomously, making it difficult to assign responsibility for their decisions. In radiology and other healthcare sectors, AI must adhere to ethical and regulatory standards to ensure fairness, safety, and transparency. A key issue is data privacy, while de-identified data is commonly used for training AI models, there is still a risk of re-identification. Regulatory frameworks must establish clear guidelines on data security to protect patient confidentiality and comply with privacy laws [\u003cspan citationid=\"CR122\" class=\"CitationRef\"\u003e122\u003c/span\u003e, \u003cspan citationid=\"CR123\" class=\"CitationRef\"\u003e123\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eBias in AI algorithms is another pressing challenge, as training datasets may not adequately represent diverse patient populations, leading to disparities in healthcare outcomes. This is particularly concerning for underserved groups, including those in rural or underfunded settings. Addressing bias requires more inclusive datasets and regulatory frameworks that promote fairness. Organizations such as the International Medical Device Regulators Forum (IMDRF) are working on global guidelines to ensure AI technologies meet high standards of safety, accuracy, and equity [\u003cspan citationid=\"CR123\" class=\"CitationRef\"\u003e123\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eBeyond improving predictive accuracy, AI is revolutionizing healthcare by enhancing the personalization of treatments. In cardiovascular imaging, for instance, AI streamlines image analysis, segmentation, and automated measurements, improving efficiency and accuracy. However, AI-driven medical devices (MLMDs) raise concerns about transparency, bias, and unforeseen behaviors, necessitating an adaptive regulatory approach. Institutions like the IMDRF advocate for a lifecycle regulatory model, including continuous post-market evaluation to monitor AI safety and efficacy [\u003cspan citationid=\"CR124\" class=\"CitationRef\"\u003e124\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eExisting regulatory frameworks, such as the Medical Device Regulation (MDR) in Europe and the FDA\u0026rsquo;s risk-based classification of medical software in the U.S., face challenges in keeping pace with AI advancements. The regulatory landscape remains fragmented, as different countries adopt varying standards, creating difficulties for multinational healthcare providers and AI developers. Global regulatory convergence is essential for streamlining AI implementation, with initiatives like the US-EU Trade and Technology Council\u0026rsquo;s AI code of conduct representing steps toward harmonization, though progress has been slow [\u003cspan citationid=\"CR121\" class=\"CitationRef\"\u003e121\u003c/span\u003e, \u003cspan citationid=\"CR125\" class=\"CitationRef\"\u003e125\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eRegulatory bodies must evaluate AI not only for predictive accuracy but also for clinical utility, decision-making impact, and patient outcomes. Prospective studies on real-world integration are particularly needed in maternal health. Adoption in resource-limited settings should consider computational demands, healthcare infrastructure, and clinician input to ensure feasibility and practicality.\u003c/p\u003e \u003cp\u003eThis systematic review demonstrates that AI tools exhibit strong internal predictive performance for key maternal and neonatal outcomes, including preterm birth, low birth weight, and neonatal mortality. However, their real-world effectiveness remains constrained by limited external validation, geographic imbalances in underlying datasets, and modest reductions in model sensitivity when applied in clinical practice. Although only a small proportion of studies reported bedside or interventional implementation, available evidence suggests that, when appropriately integrated, AI can enhance early risk stratification and support timely clinical decision-making.\u003c/p\u003e \u003cp\u003eThe findings further indicate that optimal AI performance depends on high-quality data, active clinician oversight, and robust regulatory frameworks. Effective clinical translation therefore requires addressing ethical considerations, ensuring model fairness, and developing context-specific tools, particularly in low- and middle-income countries (LMICs), where the burden of maternal and neonatal mortality is greatest.\u003c/p\u003e \u003cp\u003eOverall, these results align with the growing body of evidence supporting the potential of AI in predicting feto-maternal outcomes, while also highlighting important methodological limitations. Substantial heterogeneity, inconsistent outcome definitions, and insufficient external validation limited comparability across studies, and the review itself was constrained by the absence of formal certainty grading (e.g., GRADE). Together, these findings underscore the need for standardized methodologies, rigorous validation across diverse populations, and clearer reporting standards to better inform clinical practice and health policy.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eArtificial intelligence and machine learning hold significant promise for improving maternal and neonatal outcomes through enhanced risk prediction and decision support. However, this review underscores that clinical value depends not only on algorithmic accuracy but also on high-quality, representative data, rigorous external validation, and sustained clinician oversight. AI should function as an augmentative tool within integrated care pathways rather than a replacement for clinical judgment. Achieving real-world impact requires addressing regulatory, ethical, and operational challenges, including bias, accountability, data privacy, and contextual adaptability. Coordinated collaboration among clinicians, researchers, developers, and regulators is essential to establish transparent and equitable frameworks for deployment. With standardized methodologies and robust Governance, particularly in low- and middle-income countries where the burden is greatest. AI can progress from experimental models to trustworthy, scalable tools capable of delivering meaningful global health gains.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eGrant number\u003c/h2\u003e \u003cp\u003eNot applicable.\u003c/p\u003e \u003c/p\u003e\u003cp\u003e \u003ch2\u003eCompeting Interests\u003c/h2\u003e \u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding Statement\u003c/h2\u003e \u003cp\u003eThis research was funded by the Coordination for the Improvement of Higher Education Personnel (CAPES), Brazil, and DeepMind. The funders had no influence on the study design, data collection and analysis, interpretation of results, decision to publish, or preparation of this manuscript. There\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eODA and RCP conceptualized the study. Methodology was developed by RCP, ODA, and APMO. The search strategy was designed and implemented by APMO, ODA, and RCP. Data curation was carried out by APMO and ODA, with APMO responsible for importing records into Rayyan and, together with ODA, performing deduplication. ODA conducted the data analysis. Validation of the study was undertaken by ODA, CMC, RCP, and KEO. Supervision was provided by RCP and CT. ODA drafted the original manuscript, and all authors (ODA, CMC, APMO, KEO, CT, and RCP) contributed to critical review and editing of the manuscript. All authors approved the final version of the manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e \u003cp\u003eThis article is based on research conducted as part of my PhD thesis, which was supported by the Coordination for the Improvement of Higher Education Personnel (CAPES), Brazil, and DeepMind. The funders had no influence on the study design, data collection and analysis, interpretation of results, decision to publish, or preparation of this manuscript.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eSelected data for the included studies are provided as supplementary files. A list of excluded articles after full-text screening is also available in the supplementary material. R codes used for data handling and analysis are available from the authors upon reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eOnambele, L. et al. Trends, projections, and regional disparities of maternal mortality in Africa (1990\u0026ndash;2030): an ARIMA forecasting approach. \u003cem\u003eEpidemiologia\u003c/em\u003e \u003cb\u003e4\u003c/b\u003e, 322\u0026ndash;351 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlamrew, A. et al. Determinants of birth preparedness and complication readiness practice among reproductive-age women in Africa: a systematic review and meta-analysis. \u003cem\u003eBMC Public. Health\u003c/em\u003e. \u003cb\u003e24\u003c/b\u003e, 3154 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGreen, G. H. Trends in maternal mortality. \u003cem\u003eN Z. Med. J.\u003c/em\u003e \u003cb\u003e65\u003c/b\u003e, 80\u0026ndash;86 (1966).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBhutta, Z. A. \u0026amp; Black, R. E. Global maternal, newborn, and child health\u0026mdash;so near and yet so far. \u003cem\u003eN Engl. J. Med.\u003c/em\u003e \u003cb\u003e369\u003c/b\u003e, 2226\u0026ndash;2235. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1056/NEJMra111185\u003c/span\u003e\u003cspan address=\"10.1056/NEJMra111185\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSacks, E. et al. Comprehensive review of the evidence regarding the effectiveness of community-based primary health care in improving maternal, neonatal and child health: 3. neonatal health findings. \u003cem\u003eJ. Glob Health\u003c/em\u003e. \u003cb\u003e7\u003c/b\u003e, 010903. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.7189/jogh.07.010903\u003c/span\u003e\u003cspan address=\"10.7189/jogh.07.010903\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTobe, R. G., Islam, M. T., Yoshimura, Y. \u0026amp; Hossain Strengthening the community support group to improve maternal and neonatal health seeking behaviors: a cluster-randomized controlled trial in Satkhira District, Bangladesh. \u003cem\u003ePLoS One\u003c/em\u003e. \u003cb\u003e14\u003c/b\u003e, e0212847. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1371/journal.pone.0212847\u003c/span\u003e\u003cspan address=\"10.1371/journal.pone.0212847\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMortensen, B. et al. Improving maternal and neonatal health by a midwife-led continuity model of care\u0026mdash;an observational study in one governmental hospital in Palestine. EClinicalMedicine 10, 84\u0026ndash;91 (2019). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.eclinm.2019.04.003\u003c/span\u003e\u003cspan address=\"10.1016/j.eclinm.2019.04.003\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMajors, C. E., Smith, C. A., Natoli, M. E., Kundrod, K. A. \u0026amp; Richards-Kortum, R. Point-of-care diagnostics to improve maternal and neonatal health in low-resource settings. \u003cem\u003eLab. Chip\u003c/em\u003e. \u003cb\u003e17\u003c/b\u003e, 3351\u0026ndash;3387. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1039/C7LC00374A\u003c/span\u003e\u003cspan address=\"10.1039/C7LC00374A\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTobe, R. G., Haque, S. E., Ikegami, K. \u0026amp; Mori, R. Mobile-health tool to improve maternal and neonatal health care in Bangladesh: a cluster randomized controlled trial. \u003cem\u003eBMC Pregnancy Childbirth\u003c/em\u003e. \u003cb\u003e18\u003c/b\u003e, 102. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12884-018-1714-4\u003c/span\u003e\u003cspan address=\"10.1186/s12884-018-1714-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIlozumba, O. et al. Targeting strategies of mHealth interventions for maternal health in low and middle-income countries: a systematic review protocol. \u003cem\u003eBMJ Open.\u003c/em\u003e \u003cb\u003e8\u003c/b\u003e, e019345. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1136/bmjopen-2017-019345\u003c/span\u003e\u003cspan address=\"10.1136/bmjopen-2017-019345\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKhan, M. S., Bates, D. \u0026amp; Kovacheva, V. P. The quest for equitable health care: the potential for artificial intelligence. \u003cem\u003eNEJM Catalyst Innov. Care Deliv\u003c/em\u003e. \u003cb\u003e2\u003c/b\u003e, 14 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIftikhar, P., Kuijpers, M. V., Khayyat, A., Iftikhar, A. \u0026amp; De Sa, M. D. Artificial intelligence: a new paradigm in obstetrics and gynecology research and clinical practice. \u003cem\u003eCureus\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e, e7124. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.7759/cureus.7124\u003c/span\u003e\u003cspan address=\"10.7759/cureus.7124\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRamakrishnan, R., Rao, S. \u0026amp; He, J. R. Perinatal health predictors using artificial intelligence: a review. \u003cem\u003eWomens Health (Lond Engl)\u003c/em\u003e. \u003cb\u003e17\u003c/b\u003e, 17455065211046132. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/17455065211046132\u003c/span\u003e\u003cspan address=\"10.1177/17455065211046132\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDelanerolle, G. et al. Artificial intelligence: a rapid case for advancement in the personalization of gynaecology/obstetric and mental health care. \u003cem\u003eWomens Health (Lond Engl)\u003c/em\u003e. \u003cb\u003e17\u003c/b\u003e, 17455065211018111. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/17455065211018111\u003c/span\u003e\u003cspan address=\"10.1177/17455065211018111\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDavidson, L. \u0026amp; Boland, M. R. Towards deep phenotyping pregnancy: a systematic review on artificial intelligence and machine learning methods to improve pregnancy outcomes. \u003cem\u003eBrief. Bioinform\u003c/em\u003e. \u003cb\u003e22\u003c/b\u003e, bbaa369. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/bib/bbaa369\u003c/span\u003e\u003cspan address=\"10.1093/bib/bbaa369\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePanch, T., Szolovits, P. \u0026amp; AR. \u0026amp;. Artificial intelligence, machine learning and health systems. \u003cem\u003eJ. Glob Health\u003c/em\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.7189/jogh.08.020303\u003c/span\u003e\u003cspan address=\"10.7189/jogh.08.020303\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eObrzut, B., Kusy, M., Semczuk, A., Obrzut, M. \u0026amp; Kluska, J. Prediction of 10-year overall survival in patients with operable cervical cancer using a probabilistic neural network. \u003cem\u003eJ. Cancer\u003c/em\u003e. \u003cb\u003e10\u003c/b\u003e, 4189\u0026ndash;4195. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.7150/jca.33945\u003c/span\u003e\u003cspan address=\"10.7150/jca.33945\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRahmani, A. M. et al. ANR. Machine learning (ML) in medicine: review, applications, and challenges. Mathematics.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRutkowski, L. Springer,. Introduction to probabilistic neural networks. In Probabilistic Neural Networks 21\u0026ndash;57 (2004). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/978-3-540-40046-2_3\u003c/span\u003e\u003cspan address=\"10.1007/978-3-540-40046-2_3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSatapathy, S. K., Dehuri, S., Jagadev, A. K. \u0026amp; Mishra, S. EEG brain signal classification for epileptic seizure disorder detection. In EEG Brain Signal Classification for Epileptic Seizure Disorder Detection 1\u0026ndash;126 (2019). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/C2018-0-01888-5\u003c/span\u003e\u003cspan address=\"10.1016/C2018-0-01888-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e\u0026Ccedil;olakoğlu, N. \u0026amp; Akkaya, B. Comparison of multi-class classification algorithms on early diagnosis of heart diseases recommendation system for Spotify. In Proceedings of the Conference: Recent Advances in Data Science and Business Analytics 162\u0026ndash;171 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSharkawy, A. N. Principle of neural network and its main types (review). \u003cem\u003eJ. Adv. Appl. Comput. Math.\u003c/em\u003e \u003cb\u003e7\u003c/b\u003e, 8\u0026ndash;19. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.15377/2409-5761.2020.07.2\u003c/span\u003e\u003cspan address=\"10.15377/2409-5761.2020.07.2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang, X., Ding, Y. \u0026amp; Shao, H. The improved radial basis function neural network and its application. \u003cem\u003eArtif. Life Robot\u003c/em\u003e. \u003cb\u003e2\u003c/b\u003e, 8\u0026ndash;11. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/BF02471145\u003c/span\u003e\u003cspan address=\"10.1007/BF02471145\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (1998).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGoldenberg, R. L., McClure, E. M. \u0026amp; Saleem, S. Improving pregnancy outcomes in low- and middle-income countries. \u003cem\u003eReprod. Health\u003c/em\u003e. \u003cb\u003e15\u003c/b\u003e, 88. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12978-018-0524-5\u003c/span\u003e\u003cspan address=\"10.1186/s12978-018-0524-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBaeta, T., Rocha, A. L. L., Oliveira, J. A., Couto da Silva, A. P. \u0026amp; Reis, Z. S. N. Accuracy of machine learning and traditional statistical models in the prediction of postpartum haemorrhage: a systematic review. \u003cem\u003eBMJ Open.\u003c/em\u003e \u003cb\u003e15\u003c/b\u003e, e094455. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1136/bmjopen-2024-094455\u003c/span\u003e\u003cspan address=\"10.1136/bmjopen-2024-094455\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGravesteijn, B. Y. et al. Machine learning algorithms performed no better than regression models for prognostication in traumatic brain injury. \u003cem\u003eJ. Clin. Epidemiol.\u003c/em\u003e \u003cb\u003e122\u003c/b\u003e, 95\u0026ndash;107. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jclinepi.2020.03.005\u003c/span\u003e\u003cspan address=\"10.1016/j.jclinepi.2020.03.005\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMoher, D. et al. Preferred reporting items for systematic review and meta-analysis protocols (PRISMA-P) 2015 statement. \u003cem\u003eSyst. Rev.\u003c/em\u003e \u003cb\u003e4\u003c/b\u003e, 1. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/2046-4053-4-1\u003c/span\u003e\u003cspan address=\"10.1186/2046-4053-4-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYu, F., Liu, C. \u0026amp; Sharmin, S. Performance, usability, and user experience of Rayyan for systematic reviews. Proc. Assoc. Inf. Sci. Technol. 59, 843\u0026ndash;844 (2022). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/pra2.745\u003c/span\u003e\u003cspan address=\"10.1002/pra2.745\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMoons, K. G. M. et al. PROBAST: a tool to assess risk of bias and applicability of prediction model studies: explanation and elaboration. \u003cem\u003eAnn. Intern. Med.\u003c/em\u003e \u003cb\u003e170\u003c/b\u003e, W1\u0026ndash;W33. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.7326/M18-1377\u003c/span\u003e\u003cspan address=\"10.7326/M18-1377\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRanjbar, A. et al. Machine learning models for predicting preeclampsia: a systematic review. \u003cem\u003eBMC Pregnancy Childbirth\u003c/em\u003e. \u003cb\u003e24\u003c/b\u003e, 6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12884-023-06220-1\u003c/span\u003e\u003cspan address=\"10.1186/s12884-023-06220-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSteyerberg, E. W. \u0026amp; Harrell, F. E. Jr. Prediction models need appropriate internal, internal-external, and external validation. \u003cem\u003eJ. Clin. Epidemiol.\u003c/em\u003e \u003cb\u003e69\u003c/b\u003e, 245\u0026ndash;247. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jclinepi.2015.04.005\u003c/span\u003e\u003cspan address=\"10.1016/j.jclinepi.2015.04.005\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXu, H., Platt, R. W., Luo, Z. C., Wei, S. \u0026amp; Fraser, W. D. Exploring heterogeneity in meta-analyses: needs, resources and challenges. \u003cem\u003ePaediatr. Perinat. Epidemiol.\u003c/em\u003e \u003cb\u003e22\u003c/b\u003e, 18\u0026ndash;28. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/j.1365-3016.2007.00908.x\u003c/span\u003e\u003cspan address=\"10.1111/j.1365-3016.2007.00908.x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2008).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShen, J. et al. An innovative artificial intelligence-based app for the diagnosis of gestational diabetes mellitus (GDM-AI): development study. \u003cem\u003eJ. Med. Internet Res.\u003c/em\u003e \u003cb\u003e22\u003c/b\u003e, e21573. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2196/21573\u003c/span\u003e\u003cspan address=\"10.2196/21573\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSufriyana, H., Wu, Y. W. \u0026amp; Su, E. C. Y. Artificial intelligence-assisted prediction of preeclampsia: development and external validation of a nationwide health insurance dataset of the BPJS Kesehatan in Indonesia. EBioMedicine 54, 102710 (2020). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ebiom.2020.102710\u003c/span\u003e\u003cspan address=\"10.1016/j.ebiom.2020.102710\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTsur, A., Batsry, L., Rosenstein, M. G. \u0026amp; Barak, O. Development and validation of a machine-learning model for prediction of shoulder dystocia. \u003cem\u003eUltrasound Obstet. Gynecol.\u003c/em\u003e 588\u0026ndash;596. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/uog.21878\u003c/span\u003e\u003cspan address=\"10.1002/uog.21878\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRuppel, H. et al. Development and validation of an obstetric comorbidity risk score for clinical use. \u003cem\u003eWomens Health Rep. (New Rochelle)\u003c/em\u003e. \u003cb\u003e2\u003c/b\u003e, 507\u0026ndash;515. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1089/whr.2021.0046\u003c/span\u003e\u003cspan address=\"10.1089/whr.2021.0046\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFotiadou, E., Konopczyński, T. J. H. \u0026amp; RV. End-to-end trained encoder\u0026ndash;decoder convolutional neural network for fetal electrocardiogram signal denoising. \u003cem\u003ePhysiol. Meas.\u003c/em\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1088/1361-6579/ab69b9\u003c/span\u003e\u003cspan address=\"10.1088/1361-6579/ab69b9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHu, T. et al. Establishment of a model for predicting the outcome of induced labor in full-term pregnancy based on machine learning algorithm. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e, 19063. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41598-022-21954-2\u003c/span\u003e\u003cspan address=\"10.1038/s41598-022-21954-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eClapp, M. A., McCoy, T. H., James, K. E., Perlis, R. H. \u0026amp; Kaimal, A. J. Natural language processing of admission notes to predict postpartum hemorrhage. \u003cem\u003eAm. J. Obstet. Gynecol.\u003c/em\u003e \u003cb\u003e228\u003c/b\u003e, S534\u0026ndash;S535. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ajog.2022.11.913\u003c/span\u003e\u003cspan address=\"10.1016/j.ajog.2022.11.913\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi, Y. X. et al. Novel electronic health records applied for prediction of pre-eclampsia: machine-learning algorithms. \u003cem\u003ePregnancy Hypertens.\u003c/em\u003e \u003cb\u003e26\u003c/b\u003e, 102\u0026ndash;109. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.preghy.2021.10.006\u003c/span\u003e\u003cspan address=\"10.1016/j.preghy.2021.10.006\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eArtzi, N. S. et al. Prediction of gestational diabetes based on nationwide electronic health records. \u003cem\u003eNat. Med.\u003c/em\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41591-019-0724-8\u003c/span\u003e\u003cspan address=\"10.1038/s41591-019-0724-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi, Y. X., Liu, Y. C., Wang, M. \u0026amp; Huang, Y. L. Prediction of gestational diabetes mellitus at the first trimester: machine-learning algorithms. \u003cem\u003eArch. Gynecol. Obstet.\u003c/em\u003e \u003cb\u003e309\u003c/b\u003e, 2557\u0026ndash;2566. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00404-023-07131-4\u003c/span\u003e\u003cspan address=\"10.1007/s00404-023-07131-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShah, S. Y. et al. Prediction of postpartum hemorrhage (PPH) using machine learning algorithms in a Kenyan population. \u003cem\u003eFront. Glob Womens Health\u003c/em\u003e. \u003cb\u003e4\u003c/b\u003e, 1161157. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fgwh.2023.1161157\u003c/span\u003e\u003cspan address=\"10.3389/fgwh.2023.1161157\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLipschuetz, M. et al. Prediction of vaginal birth after cesarean deliveries using machine learning. \u003cem\u003eAm. J. Obstet. Gynecol.\u003c/em\u003e \u003cb\u003e222\u003c/b\u003e, 613e1\u0026ndash;613e12. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ajog.2019.12.267\u003c/span\u003e\u003cspan address=\"10.1016/j.ajog.2019.12.267\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi, S. et al. Improving preeclampsia risk prediction by modeling pregnancy trajectories from routinely collected electronic medical record data. \u003cem\u003eNPJ Digit. Med.\u003c/em\u003e \u003cb\u003e5\u003c/b\u003e, 68. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41746-022-00612-x\u003c/span\u003e\u003cspan address=\"10.1038/s41746-022-00612-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eQiu, J. Fetal movement detection by wearable accelerometer duo based on machine learning. \u003cem\u003eIEEE Sens. J.\u003c/em\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/JSEN.2022.3172451\u003c/span\u003e\u003cspan address=\"10.1109/JSEN.2022.3172451\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDouali, N., Dollon, J. \u0026amp; Jaulent, M. C. Personalized prediction of gestational diabetes using a clinical decision support system. In IEEE International Conference on Fuzzy Systems (FUZZ-IEEE) 1\u0026ndash;5 (2015). (2015). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/FUZZ-IEEE.2015.7337813\u003c/span\u003e\u003cspan address=\"10.1109/FUZZ-IEEE.2015.7337813\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGu, X. et al. Machine learning approach for the prediction of macrosomia. \u003cem\u003eVis. Comput. Ind. Biomed. Art\u003c/em\u003e. \u003cb\u003e7\u003c/b\u003e, 22. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s42492-024-00172-9\u003c/span\u003e\u003cspan address=\"10.1186/s42492-024-00172-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang, J. et al. Machine learning-based risk stratification for gestational diabetes management. \u003cem\u003eSensors\u003c/em\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/s22134805\u003c/span\u003e\u003cspan address=\"10.3390/s22134805\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStarling, S. A molecular signal for preeclampsia. \u003cem\u003eNat. Rev. Endocrinol.\u003c/em\u003e \u003cb\u003e16\u003c/b\u003e, 471\u0026ndash;472 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAhmed, M., Kashem, M. A., Rahman, M. \u0026amp; Khatun, S. Review and analysis of risk factor of maternal health in remote area using the internet of things (IoT). In Lecture Notes in Computer Science, vol. 632, 1\u0026ndash;10 (Springer, Singapore, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/978-981-15-2317-5_30\u003c/span\u003e\u003cspan address=\"10.1007/978-981-15-2317-5_30\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang, N. et al. Development and validation of risk prediction models for gestational diabetes mellitus using four different methods. \u003cem\u003eMetabolites\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e, 1040. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/metabo12111040\u003c/span\u003e\u003cspan address=\"10.3390/metabo12111040\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMazloom, A. R. et al. Noninvasive prenatal detection of sex chromosomal aneuploidies by sequencing circulating cell-free DNA from maternal plasma. \u003cem\u003ePrenat Diagn.\u003c/em\u003e \u003cb\u003e33\u003c/b\u003e, 591\u0026ndash;597. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/pd.4127\u003c/span\u003e\u003cspan address=\"10.1002/pd.4127\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHoffman, M. K., Ma, N. \u0026amp; Roberts, A. A machine learning algorithm for predicting maternal readmission for hypertensive disorders of pregnancy. \u003cem\u003eAm. J. Obstet. Gynecol. MFM\u003c/em\u003e. \u003cb\u003e3\u003c/b\u003e, 100250. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ajogmf.2020.100250\u003c/span\u003e\u003cspan address=\"10.1016/j.ajogmf.2020.100250\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePayne, J. L. et al. DNA methylation biomarkers prospectively predict both antenatal and postpartum depression. \u003cem\u003ePsychiatry Res.\u003c/em\u003e \u003cb\u003e285\u003c/b\u003e, 112711. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.psychres.2019.112711\u003c/span\u003e\u003cspan address=\"10.1016/j.psychres.2019.112711\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVenkatesh, K. K. et al. Machine learning and statistical models to predict postpartum hemorrhage. \u003cem\u003eObstet. Gynecol.\u003c/em\u003e \u003cb\u003e135\u003c/b\u003e, 935\u0026ndash;944. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1097/AOG.0000000000003759\u003c/span\u003e\u003cspan address=\"10.1097/AOG.0000000000003759\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhong, Q. Y. et al. Use of natural language processing in electronic medical records to identify pregnant women with suicidal behavior: towards a solution to the complex classification problem. \u003cem\u003eEur. J. Epidemiol.\u003c/em\u003e \u003cb\u003e34\u003c/b\u003e, 153\u0026ndash;162. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10654-018-0470-0\u003c/span\u003e\u003cspan address=\"10.1007/s10654-018-0470-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBadr, D. A., De Lucia, F., Carlin, A., Jani, J. C., Cannie, M. M. \u0026amp; Chest CT scan predictors of intensive care unit admission in hospitalized pregnant women with COVID-19: a case-control study. \u003cem\u003eJ. Matern Fetal Neonatal Med.\u003c/em\u003e \u003cb\u003e36\u003c/b\u003e, 2241107. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/14767058.2023.2241107\u003c/span\u003e\u003cspan address=\"10.1080/14767058.2023.2241107\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLeonard, S. A. et al. Obstetric comorbidity scores and disparities in severe maternal morbidity across marginalized groups. \u003cem\u003eAm. J. Obstet. Gynecol. MFM\u003c/em\u003e. \u003cb\u003e4\u003c/b\u003e, 100530. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ajogmf.2021.100530\u003c/span\u003e\u003cspan address=\"10.1016/j.ajogmf.2021.100530\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDuval, A. et al. A hybrid artificial intelligence model leverages multi-centric clinical data to improve fetal heart rate pregnancy prediction across time-lapse systems. \u003cem\u003eHum. Reprod.\u003c/em\u003e \u003cb\u003e38\u003c/b\u003e, 596\u0026ndash;608. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/humrep/dead023\u003c/span\u003e\u003cspan address=\"10.1093/humrep/dead023\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuedalia, J. et al. Transporting an artificial intelligence model to predict emergency cesarean delivery: overcoming challenges posed by interfacility variation. \u003cem\u003eJ. Med. Internet Res.\u003c/em\u003e \u003cb\u003e23\u003c/b\u003e, e28120. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2196/28120\u003c/span\u003e\u003cspan address=\"10.2196/28120\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKrupa, J. D., Dhanalakshmi, A., R, K. \u0026amp; S. \u0026amp; An improved parallel sub-filter adaptive noise canceler for the extraction of fetal ECG. \u003cem\u003eBiomed. Tech. (Berl)\u003c/em\u003e. \u003cb\u003e66\u003c/b\u003e, 503\u0026ndash;514. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1515/bmt-2020-0313\u003c/span\u003e\u003cspan address=\"10.1515/bmt-2020-0313\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMateen, B. A., David, A. L. \u0026amp; Denaxas, S. Electronic health records to predict gestational diabetes risk. \u003cem\u003eTrends Pharmacol. Sci.\u003c/em\u003e \u003cb\u003e41\u003c/b\u003e, 301\u0026ndash;304. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.tips.2020.03.003\u003c/span\u003e\u003cspan address=\"10.1016/j.tips.2020.03.003\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePark, T. J. et al. Machine learning model for classifying the results of fetal cardiotocography conducted in high-risk pregnancies. \u003cem\u003eYonsei Med. J.\u003c/em\u003e \u003cb\u003e63\u003c/b\u003e, 692\u0026ndash;700. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3349/ymj.2022.63.7.692\u003c/span\u003e\u003cspan address=\"10.3349/ymj.2022.63.7.692\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePinto, Y. et al. Gestational diabetes is driven by microbiota-induced inflammation months before diagnosis. \u003cem\u003eGut\u003c/em\u003e \u003cb\u003e72\u003c/b\u003e, 918\u0026ndash;928. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1136/gutjnl-2022-328406\u003c/span\u003e\u003cspan address=\"10.1136/gutjnl-2022-328406\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eQiao, S. et al. FLDS: An intelligent feature learning detection system for visualizing medical images supporting fetal four-chamber views. \u003cem\u003eIEEE J. Biomed. Health Inf.\u003c/em\u003e \u003cb\u003e26\u003c/b\u003e, 4814\u0026ndash;4825. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/JBHI.2021.3091579\u003c/span\u003e\u003cspan address=\"10.1109/JBHI.2021.3091579\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAndreasen, L. A. et al. Multi-centre deep learning for placenta segmentation in obstetric ultrasound with multi-observer and cross-country generalization. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e, 2221. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41598-023-29105-x\u003c/span\u003e\u003cspan address=\"10.1038/s41598-023-29105-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBonet-Carne, E. et al. Quantitative ultrasound texture analysis of fetal lungs to predict neonatal respiratory morbidity. \u003cem\u003eUltrasound Obstet. Gynecol.\u003c/em\u003e \u003cb\u003e45\u003c/b\u003e, 427\u0026ndash;433. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/uog.13441\u003c/span\u003e\u003cspan address=\"10.1002/uog.13441\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlzubaidi, M. et al. Ensemble transfer learning for fetal head analysis: from segmentation to gestational age and weight prediction. \u003cem\u003eDiagnostics (Basel)\u003c/em\u003e. \u003cb\u003e12\u003c/b\u003e, 1\u0026ndash;28. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/diagnostics12092229\u003c/span\u003e\u003cspan address=\"10.3390/diagnostics12092229\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLeonard, S. A., Kennedy, C. J., Carmichael, S. L., Lyell, D. J. \u0026amp; Main, E. K. An expanded obstetric comorbidity scoring system for predicting severe maternal morbidity. \u003cem\u003eObstet. Gynecol.\u003c/em\u003e \u003cb\u003e136\u003c/b\u003e, 440\u0026ndash;449. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1097/AOG.0000000000004022\u003c/span\u003e\u003cspan address=\"10.1097/AOG.0000000000004022\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAfrizal, S. H. et al. A user-centered design of natural language processing for maternal monitoring chatbot system. In 2022 International Conference on Informatics, Multimedia, Cyber and Information System (ICIMCIS) 244\u0026ndash;248 (2022). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/ICIMCIS56303.2022.10017517\u003c/span\u003e\u003cspan address=\"10.1109/ICIMCIS56303.2022.10017517\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCaballero-Ruiz, E. et al. A web-based clinical decision support system for gestational diabetes: automatic diet prescription and detection of insulin needs. \u003cem\u003eInt. J. Med. Inf.\u003c/em\u003e \u003cb\u003e102\u003c/b\u003e, 35\u0026ndash;49. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ijmedinf.2017.02.014\u003c/span\u003e\u003cspan address=\"10.1016/j.ijmedinf.2017.02.014\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDuysburgh, E. et al. Quality of antenatal and childbirth care in rural health facilities in Burkina Faso, Ghana and Tanzania: an intervention study. \u003cem\u003eTrop. Med. Int. Health\u003c/em\u003e. \u003cb\u003e21\u003c/b\u003e, 70\u0026ndash;83. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/tmi.12627\u003c/span\u003e\u003cspan address=\"10.1111/tmi.12627\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBrocklehurst, P. et al. Computerised interpretation of fetal heart rate during labour (INFANT): a randomised controlled trial. \u003cem\u003eLancet\u003c/em\u003e \u003cb\u003e389\u003c/b\u003e, 1719\u0026ndash;1729. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/S0140-6736(17)30568-8\u003c/span\u003e\u003cspan address=\"10.1016/S0140-6736(17)30568-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMcDonald, S. C., Brooker, G., Phipps, H. \u0026amp; Hyett, J. The identification and tracking of uterine contractions using template based cross-correlation. \u003cem\u003eAnn. Biomed. Eng.\u003c/em\u003e \u003cb\u003e45\u003c/b\u003e, 2196\u0026ndash;2210. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10439-017-1873-x\u003c/span\u003e\u003cspan address=\"10.1007/s10439-017-1873-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCarroll, A. E., Biondich, P., Anand, V., Dugan, T. M. \u0026amp; Downs, S. M. A randomized controlled trial of screening for maternal depression with a clinical decision support system. \u003cem\u003eJ. Am. Med. Inf. Assoc.\u003c/em\u003e \u003cb\u003e20\u003c/b\u003e, 311\u0026ndash;316. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1136/amiajnl-2011-000682\u003c/span\u003e\u003cspan address=\"10.1136/amiajnl-2011-000682\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlabi, O. et al. Robust fetoscopic mosaicking from deep learned flow fields. \u003cem\u003eInt. J. Comput. Assist. Radiol. Surg.\u003c/em\u003e \u003cb\u003e17\u003c/b\u003e, 1125\u0026ndash;1134. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11548-022-02623-1\u003c/span\u003e\u003cspan address=\"10.1007/s11548-022-02623-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, Y. et al. Therapeutic effect of cervical cerclage on cervical insufficiency during pregnancy analyzed by magnetic resonance image under neural network algorithm. \u003cem\u003eSci. Program.\u003c/em\u003e 1\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1155/2021/5879255\u003c/span\u003e\u003cspan address=\"10.1155/2021/5879255\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMarques, J. A. L. et al. IoT-based smart health system for ambulatory maternal and fetal monitoring. \u003cem\u003eIEEE Internet Things J.\u003c/em\u003e \u003cb\u003e8\u003c/b\u003e, 16814\u0026ndash;16824. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/JIOT.2020.3037759\u003c/span\u003e\u003cspan address=\"10.1109/JIOT.2020.3037759\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHawken, S. et al. External validation of machine learning models including newborn metabolomic markers for postnatal gestational age estimation in East and South-East Asian populations. \u003cem\u003eGates Open. Res.\u003c/em\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSendra-Balcells, C. et al. Generalisability of fetal ultrasound deep learning models to low-resource imaging settings in five African countries. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e, 1\u0026ndash;10. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41598-023-29490-3\u003c/span\u003e\u003cspan address=\"10.1038/s41598-023-29490-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEscobar, G. J. et al. Prediction of obstetrical and fetal complications using automated electronic health record data. \u003cem\u003eAm. J. Obstet. Gynecol.\u003c/em\u003e \u003cb\u003e224\u003c/b\u003e, 137\u0026ndash;147e7. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ajog.2020.10.030\u003c/span\u003e\u003cspan address=\"10.1016/j.ajog.2020.10.030\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNeves, M. R. et al. Causal dynamic Bayesian networks for the management of glucose control in gestational diabetes. In. IEEE 9th International Conference on Healthcare Informatics (ICHI) 31\u0026ndash;40 (2021). (2021). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/ICHI52183.2021.00018\u003c/span\u003e\u003cspan address=\"10.1109/ICHI52183.2021.00018\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOliver, A. S. \u0026amp; Ashwanthika, U. A. R. Labor prediction in pregnant woman based on electromyogram and electrohysterogram using machine learning techniques. \u003cem\u003eIEEE ICISC\u003c/em\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEspinosa, C. A. et al. Multiomic signals associated with maternal epidemiological factors contributing to preterm birth in low- and middle-income countries. \u003cem\u003eSci. Adv.\u003c/em\u003e \u003cb\u003e9\u003c/b\u003e, eade7692. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1126/sciadv.ade7692\u003c/span\u003e\u003cspan address=\"10.1126/sciadv.ade7692\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNaimi, A. I., Platt, R. W. \u0026amp; Larkin, J. C. Machine learning for fetal growth prediction. \u003cem\u003eEpidemiology\u003c/em\u003e \u003cb\u003e29\u003c/b\u003e, 290\u0026ndash;298. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1097/EDE.0000000000000788\u003c/span\u003e\u003cspan address=\"10.1097/EDE.0000000000000788\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJoglekar, M. V. et al. Postpartum circulating microRNA enhances prediction of future type 2 diabetes in women with previous gestational diabetes. \u003cem\u003eDiabetologia\u003c/em\u003e \u003cb\u003e64\u003c/b\u003e, 1516\u0026ndash;1526. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00125-021-05429-z\u003c/span\u003e\u003cspan address=\"10.1007/s00125-021-05429-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEdelman, E. A. et al. Implementation of an electronic genomic and family health history tool in primary prenatal care. \u003cem\u003eAm. J. Med. Genet. C Semin Med. Genet.\u003c/em\u003e \u003cb\u003e166C\u003c/b\u003e, 34\u0026ndash;44. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/ajmg.c.31389\u003c/span\u003e\u003cspan address=\"10.1002/ajmg.c.31389\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLowery, D. et al. OB HUB: Remote electronic fetal monitoring surveillance. \u003cem\u003eMCN Am. J. Matern Child. Nurs.\u003c/em\u003e \u003cb\u003e48\u003c/b\u003e, 62\u0026ndash;68. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1097/NMC.0000000000000891\u003c/span\u003e\u003cspan address=\"10.1097/NMC.0000000000000891\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLuo, D. et al. A prenatal ultrasound scanning approach: one-touch technique in second and third trimesters. \u003cem\u003eUltrasound Med. Biol.\u003c/em\u003e \u003cb\u003e47\u003c/b\u003e, 2258\u0026ndash;2265. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ultrasmedbio.2021.04.020\u003c/span\u003e\u003cspan address=\"10.1016/j.ultrasmedbio.2021.04.020\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMa, M. et al. Diagnostic performance of fetal intelligent navigation echocardiography (FINE) in fetuses with double-outlet right ventricle (DORV). \u003cem\u003eInt. J. Cardiovasc. Imaging\u003c/em\u003e. \u003cb\u003e36\u003c/b\u003e, 2165\u0026ndash;2172. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10554-020-01932-3\u003c/span\u003e\u003cspan address=\"10.1007/s10554-020-01932-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOcak, H. A medical decision support system based on support vector machines and the genetic algorithm for the evaluation of fetal well-being. \u003cem\u003eJ. Med. Syst.\u003c/em\u003e \u003cb\u003e37\u003c/b\u003e, 9913. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10916-012-9913-4\u003c/span\u003e\u003cspan address=\"10.1007/s10916-012-9913-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePerez-Gonzalez, J., Ar\u0026aacute;mbula Cos\u0026iacute;o, F. \u0026amp; Huegel, J. C. \u0026amp; Medina-Ba\u0026ntilde;uelos, V. Probabilistic learning coherent point drift for 3D ultrasound fetal head registration. Comput. Math. Methods Med. 4271519 (2020). (2020). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1155/2020/4271519\u003c/span\u003e\u003cspan address=\"10.1155/2020/4271519\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePluym, I. D. et al. Accuracy of automated three-dimensional ultrasound imaging technique for fetal head biometry. \u003cem\u003eUltrasound Obstet. Gynecol.\u003c/em\u003e \u003cb\u003e57\u003c/b\u003e, 798\u0026ndash;803. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/uog.22171\u003c/span\u003e\u003cspan address=\"10.1002/uog.22171\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRigla, M., Mart\u0026iacute;nez-Sarriegui, I., Garc\u0026iacute;a-S\u0026aacute;ez, G., Pons, B. \u0026amp; Hernando, M. E. Gestational diabetes management using smart mobile telemedicine. \u003cem\u003eJ. Diabetes Sci. Technol.\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e, 260\u0026ndash;264. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/1932296817704442\u003c/span\u003e\u003cspan address=\"10.1177/1932296817704442\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZheng, X., Li, X., Xu, J. \u0026amp; Wei, Y. Intelligent recognition algorithm-based color Doppler ultrasound in the treatment of dangerous placenta previa. J. Healthc. Eng. 9886521 (2021). (2021). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1155/2021/9886521\u003c/span\u003e\u003cspan address=\"10.1155/2021/9886521\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFung, R. et al. International Fetal and Newborn Growth Consortium for the 21st Century (INTERGROWTH-21st). Achieving accurate estimates of fetal gestational age and personalised predictions of fetal growth based on data from an international prospective cohort study: a population-based machine learning study. \u003cem\u003eLancet Digit. Health\u003c/em\u003e. \u003cb\u003e2\u003c/b\u003e, e368\u0026ndash;e375. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/S2589-7500(20)30131-X\u003c/span\u003e\u003cspan address=\"10.1016/S2589-7500(20)30131-X\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMontenegro, J. L. Z., da Costa, C. A. \u0026amp; Janssen, L. P. Evaluating the use of chatbot during pregnancy: a usability study. \u003cem\u003eHealthc. Anal. (N Y)\u003c/em\u003e. \u003cb\u003e2\u003c/b\u003e, 100072. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.health.2022.100072\u003c/span\u003e\u003cspan address=\"10.1016/j.health.2022.100072\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu, L. et al. Fetal ultrasound image quality assessment with deep convolutional networks. \u003cem\u003eIEEE Trans. Cybern\u003c/em\u003e. \u003cb\u003e47\u003c/b\u003e, 1336\u0026ndash;1349. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/TCYB.2017.2671898\u003c/span\u003e\u003cspan address=\"10.1109/TCYB.2017.2671898\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCarlisle, N. et al. Clinicians\u0026rsquo; experiences of using and implementing a medical mobile phone app (QUiPP V2) designed to predict the risk of preterm birth and aid clinical decision making. \u003cem\u003eBMC Med. Inf. Decis. Mak.\u003c/em\u003e \u003cb\u003e21\u003c/b\u003e, 320. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12911-021-01681-w\u003c/span\u003e\u003cspan address=\"10.1186/s12911-021-01681-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChuang, L., Hwang, J. Y., Chang, C. H., Yu, C. H. \u0026amp; Chang, F. M. Ultrasound estimation of fetal weight with the use of computerized artificial neural network model. \u003cem\u003eUltrasound Med. Biol.\u003c/em\u003e \u003cb\u003e28\u003c/b\u003e, 991\u0026ndash;996. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/S0301-5629(02)00554-9\u003c/span\u003e\u003cspan address=\"10.1016/S0301-5629(02)00554-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2002).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFlaucher, M. et al. Smartphone-based colorimetric analysis of urine test strips for at-home prenatal care. \u003cem\u003eIEEE J. Transl Eng. Health Med.\u003c/em\u003e \u003cb\u003e10\u003c/b\u003e, 2800109. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/JTEHM.2022.3179147\u003c/span\u003e\u003cspan address=\"10.1109/JTEHM.2022.3179147\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHernando, M. E., G\u0026oacute;mez, E. J., Corcoy, R. \u0026amp; del Pozo, F. Evaluation of DIABNET, a decision support system for therapy planning in gestational diabetes. \u003cem\u003eComput. Methods Programs Biomed.\u003c/em\u003e \u003cb\u003e62\u003c/b\u003e, 235\u0026ndash;248. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/S0169-2607(00)00070-5\u003c/span\u003e\u003cspan address=\"10.1016/S0169-2607(00)00070-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2000).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePietrolucci, M. E. et al. Evaluation of an artificial intelligent algorithm (Heartassist\u0026trade;) to automatically assess the quality of second trimester cardiac views: a prospective study. \u003cem\u003eJ. Perinat. Med.\u003c/em\u003e \u003cb\u003e51\u003c/b\u003e, 920\u0026ndash;924. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1515/jpm-2023-0052\u003c/span\u003e\u003cspan address=\"10.1515/jpm-2023-0052\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRamspek, C. L., Jager, K. J., Dekker, F. W., Zoccali, C. \u0026amp; van Diepen, M. External validation of prognostic models: what, why, how, when and where? Clin. \u003cem\u003eKidney J.\u003c/em\u003e \u003cb\u003e14\u003c/b\u003e, 49\u0026ndash;58. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/ckj/sfaa188\u003c/span\u003e\u003cspan address=\"10.1093/ckj/sfaa188\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMachine Learning and the Profession of Medicine. \u003cem\u003eJAMA\u003c/em\u003e \u003cb\u003e316\u003c/b\u003e, 4\u0026ndash;5 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGulshan, V. et al. Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs. \u003cem\u003eJAMA\u003c/em\u003e \u003cb\u003e316\u003c/b\u003e, 2402\u0026ndash;2410. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1001/jama.2016.17216\u003c/span\u003e\u003cspan address=\"10.1001/jama.2016.17216\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDiagnostic Assessment of Deep Learning Algorithms for Detection of Lymph Node Metastases in Women With Breast Cancer. \u003cem\u003eJAMA\u003c/em\u003e \u003cb\u003e318\u003c/b\u003e, 2199\u0026ndash;2210 \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1001/jama.2017.14585\u003c/span\u003e\u003cspan address=\"10.1001/jama.2017.14585\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTopol, E. J. High-performance medicine: the convergence of human and artificial intelligence. \u003cem\u003eNat. Med.\u003c/em\u003e \u003cb\u003e25\u003c/b\u003e, 44\u0026ndash;56. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41591-018-0300-7\u003c/span\u003e\u003cspan address=\"10.1038/s41591-018-0300-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSteele, A. J., Denaxas, S. C., Shah, A. D., Hemingway, H. \u0026amp; Luscombe, N. M. Machine learning models in electronic health records can outperform conventional survival models for predicting patient mortality in coronary artery disease. \u003cem\u003ePLoS One\u003c/em\u003e. \u003cb\u003e13\u003c/b\u003e, e0202344. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1371/journal.pone.0202344\u003c/span\u003e\u003cspan address=\"10.1371/journal.pone.0202344\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRajula, H. S. R., Mauri, M. \u0026amp; Fanos, V. Scale-free networks in metabolomics. \u003cem\u003eBioinformation\u003c/em\u003e \u003cb\u003e14\u003c/b\u003e, 140\u0026ndash;144. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.6026/97320630014140\u003c/span\u003e\u003cspan address=\"10.6026/97320630014140\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRajula, H. S. R., Verlato, G., Manchia, M., Antonucci, N. \u0026amp; Fanos, V. Comparison of conventional statistical methods with machine learning in medicine: diagnosis, drug development, and treatment. \u003cem\u003eMed. (Lithuania)\u003c/em\u003e. \u003cb\u003e56\u003c/b\u003e, 455. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/medicina56090455\u003c/span\u003e\u003cspan address=\"10.3390/medicina56090455\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEsteva, A. et al. Dermatologist-level classification of skin cancer with deep neural networks. \u003cem\u003eNature\u003c/em\u003e \u003cb\u003e542\u003c/b\u003e, 115\u0026ndash;118. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/nature21056\u003c/span\u003e\u003cspan address=\"10.1038/nature21056\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBarnett-Itzhaki, Z. et al. Machine learning vs. classic statistics for the prediction of IVF outcomes. \u003cem\u003eJ. Assist. Reprod. Genet.\u003c/em\u003e \u003cb\u003e37\u003c/b\u003e, 2405\u0026ndash;2412. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10815-020-01908-1\u003c/span\u003e\u003cspan address=\"10.1007/s10815-020-01908-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShin, S. et al. Machine learning vs. conventional statistical models for predicting heart failure readmission and mortality. \u003cem\u003eESC Heart Fail.\u003c/em\u003e \u003cb\u003e8\u003c/b\u003e, 106\u0026ndash;115. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/ehf2.13073\u003c/span\u003e\u003cspan address=\"10.1002/ehf2.13073\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSong, X., Arnold, M., Jafna, C. \u0026amp; Kenneth, R. Comparison of machine learning techniques with classical statistical models in predicting health outcomes. (2004). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3233/978-1-60750-949-3-736\u003c/span\u003e\u003cspan address=\"10.3233/978-1-60750-949-3-736\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKuhle, S. et al. Comparison of logistic regression with machine learning methods for the prediction of fetal growth abnormalities: a retrospective cohort study. \u003cem\u003eBMC Pregnancy Childbirth\u003c/em\u003e. \u003cb\u003e18\u003c/b\u003e, 333. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12884-018-1971-2\u003c/span\u003e\u003cspan address=\"10.1186/s12884-018-1971-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDong, X. L. \u0026amp; Rekatsinas, T. Data integration and machine learning: a natural synergy. In Proceedings of the ACM SIGMOD International Conference on Management of Data 1645\u0026ndash;1650 (2018). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1145/3183713.3197387\u003c/span\u003e\u003cspan address=\"10.1145/3183713.3197387\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eParisineni, S. R. A. \u0026amp; Pal, M. Enhancing trust and interpretability of complex machine learning models using local interpretable model agnostic SHAP explanations. \u003cem\u003eInt. J. Data Sci. Anal.\u003c/em\u003e \u003cb\u003e18\u003c/b\u003e, 457\u0026ndash;466. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s41060-023-00458-w\u003c/span\u003e\u003cspan address=\"10.1007/s41060-023-00458-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKalusivalingam, A. K., Sharma, A., Patel, N. \u0026amp; Singh, V. Leveraging SHAP and LIME for enhanced explainability in AI-driven diagnostic systems. 1\u0026ndash;23 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePalaniappan, K., Lin, E. Y. T., Vogel, S. \u0026amp; Lim, J. C. W. Gaps in the global regulatory frameworks for the use of artificial intelligence (AI) in the healthcare services sector and key recommendations. \u003cem\u003eHealthc. (Basel)\u003c/em\u003e. \u003cb\u003e12\u003c/b\u003e, 1730. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/healthcare12171730\u003c/span\u003e\u003cspan address=\"10.3390/healthcare12171730\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHarvey, H. B. \u0026amp; Gowda, V. Regulatory issues and challenges to artificial intelligence adoption. \u003cem\u003eRadiol. Clin. North. Am.\u003c/em\u003e \u003cb\u003e59\u003c/b\u003e, 1075\u0026ndash;1083. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.rcl.2021.07.007\u003c/span\u003e\u003cspan address=\"10.1016/j.rcl.2021.07.007\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePesapane, F., Volont\u0026eacute;, C., Codari, M. \u0026amp; Sardanelli, F. Artificial intelligence as a medical device in radiology: ethical and regulatory issues in Europe and the United States. \u003cem\u003eInsights Imaging\u003c/em\u003e. \u003cb\u003e9\u003c/b\u003e, 745\u0026ndash;753. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s13244-018-0645-y\u003c/span\u003e\u003cspan address=\"10.1007/s13244-018-0645-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWellnhofer, E. Real-world and regulatory perspectives of artificial intelligence in cardiovascular imaging. \u003cem\u003eFront. Cardiovasc. Med.\u003c/em\u003e \u003cb\u003e9\u003c/b\u003e, 890809. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fcvm.2022.890809\u003c/span\u003e\u003cspan address=\"10.3389/fcvm.2022.890809\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYaeger, K. A., Martini, M., Yaniv, G., Oermann, E. K. \u0026amp; Costa, A. B. United States regulatory approval of medical devices and software applications enhanced by artificial intelligence. \u003cem\u003eHealth Policy Technol.\u003c/em\u003e \u003cb\u003e8\u003c/b\u003e, 192\u0026ndash;197. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.hlpt.2019.05.006\u003c/span\u003e\u003cspan address=\"10.1016/j.hlpt.2019.05.006\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2019).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Artificial intelligence, Maternal health, Neonatal outcomes, Machine learning, Predictive modeling, Systematic review","lastPublishedDoi":"10.21203/rs.3.rs-8651301/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8651301/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eMaternal mortality remains a major global challenge, especially in developing countries. This review assessed Artificial Intelligence applications in Feto-Maternal health, focusing on validation, performance, and implementation. Following PRISMA guidelines and PROSPERO registration (CRD42023347209), we searched PubMed, EMBASE, Cochrane Library, and Web of Science for studies (2000\u0026ndash;2023) applying AI to maternal or neonatal outcomes with defined methodology and validation. Of 14,049 studies, 72 met inclusion criteria; 38 (52.8%) reported external validation, and 11 (15.3%) involved experimental or interventional use. Most datasets (81.3%) were from high-income countries, mainly Asia and North America. Frequently assessed outcomes included preterm birth, low birth weight, and neonatal mortality. Random Forest and XGBoost were most used. Internal performance was strong (AUC 0.82; accuracy 89%), with modest declines in external validation (AUC 0.80; accuracy 87%) and reduced sensitivity in real-world settings. AI shows promise but requires rigorously validated, context-specific models, ethical oversight, and readiness for LMIC integration.\u003c/p\u003e","manuscriptTitle":"Artificial Intelligence in Feto-maternal Health: A Systematic Review of Predictive Models, Validation, and Clinical Translation","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-20 01:21:31","doi":"10.21203/rs.3.rs-8651301/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-03-30T09:51:06+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-22T09:26:53+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-18T02:42:09+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"325482609226908598524196779283945304495","date":"2026-03-17T08:41:47+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"267913309663906366516514074508545135559","date":"2026-03-17T03:37:36+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-03-16T09:22:41+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-01-23T04:53:40+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-01-21T06:48:58+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-01-21T06:46:01+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2026-01-20T15:27:01+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"8c64b4b6-9073-431f-9a79-bda0238af801","owner":[],"postedDate":"March 20th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":64810805,"name":"Health sciences/Diseases"},{"id":64810806,"name":"Health sciences/Health care"},{"id":64810807,"name":"Health sciences/Medical research"}],"tags":[],"updatedAt":"2026-04-22T09:56:15+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-20 01:21:31","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8651301","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8651301","identity":"rs-8651301","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}