Development and validation of a machine-learning model for prediction of shoulder dystocia

Can shoulder dystocia be predicted before operative vaginal delivery using a score that includes ultrasonographic head-perineum distance measurement?

OBJECTIVE

The main study objective was to develop a novel shoulder dystocia (SD) prediction score using ultrasound-based head-perineum distance measured before an operative vaginal delivery (OVD).

METHODS

This retrospective unicentric study (Lille, France) included all cases of OVD of singleton pregnancies from March 2019 to October 2020, with cephalic presentation and > 37 weeks of gestation, for which intrapartum sonography was performed. A multiclass-penalized logistic regression model was used to develop the SD prognostic score, with missing values imputed by multiple imputations.

RESULTS

Among the 1708 patients with OVD, 773 who underwent ultrasound for head-perineum distance were included. SD occurred in 99 cases (12.8%). The SD's predicting factors (and their weights) included the following: maternal age younger than 28 years (3 points); multiparous (4 points); induced labor (4 points); gestational diabetes (3 points); and head-perineum distance without pressure (≤20 mm [-2 points], using 21-30 mm as reference, 31-40 mm [2 points], 41-50 mm [4 points], 51-60 mm [6 points], and >60 mm [8 points]). Three patient risk subgroups were categorized as score range (occurrence percentage) as low risk: < 3 (< 10%), high risk: 3-8 (10%-20%), and very high risk: > 8 (> 20%).

CONCLUSION

The developed scoring system may help predict SD occurrence during OVD using five delivery room parameters. Replication with other populations and prospective cohorts will be needed for validation.

Association of magnetic resonance imaging-derived maternal and fetal parameters with shoulder dystocia: matched case-control study

OBJECTIVE

To assess the association of fetal body measurements and maternal pelvic measurements obtained using magnetic resonance imaging (MRI) with the incidence of shoulder dystocia.

METHODS

This was a retrospective, single-center, case-control study conducted between January 2015 and December 2022. Patients whose delivery was complicated by shoulder dystocia and who underwent fetal MRI in the third trimester were included in the case group. Patients without shoulder dystocia who were delivered normally and who also underwent fetal MRI in the third trimester were included in the control group. Cases of multiple pregnancy, planned or emergency Cesarean delivery, fetal malformation or those with incomplete MRI examination were excluded. The case group was matched with the control group in a 1:2 ratio according to maternal age, maternal body mass index, gestational diabetes mellitus, diabetes mellitus Type 1 or 2, gestational age at MRI examination, gestational age at birth and birth weight. Shoulder dystocia was defined as per the Royal College of Obstetricians and Gynecologists and significant shoulder dystocia was defined as shoulder dystocia that was not resolved by the McRoberts' maneuver or suprapubic pressure. The following fetal and maternal measurements were quantified on MRI in both groups by two readers (one experienced and one inexperienced physician) who were blinded to the obstetric outcomes: fetal body volume (FBV), shoulder skin-to-skin distance, interhumeral distance, biparietal diameter (BPD), head circumference, obstetric conjugate (OC), sagittal outlet diameter (SOD), coccygeal pelvic outlet (CPO) and maximal transverse diameter (MTD). A stepwise backward logistic regression that included all measurements was performed. The inter-rater reliability of the measurements was estimated using interclass correlation coefficient (ICC). Statistical significance was set at P < 0.05.

RESULTS

Among the 1843 patients included in the study, there were 63 (3.4%) cases of shoulder dystocia. After matching, the case group comprised 36 patients and the control group comprised 72 patients. Patients who had shoulder dystocia, compared to those without, had higher FBV (P = 0.023), higher shoulder skin-to-skin distance (P = 0.003), lower OC (P = 0.021), lower SOD (P = 0.004), lower CPO (P = 0.045) and lower MTD (P = 0.001) in comparison with those without. The logistic regression model showed that FBV, shoulder skin-to-skin distance, BPD, SOD and MTD were independent predictors of shoulder dystocia. The measurements of interest had moderate to excellent reliability when repeated by an inexperienced reader. In those who had non-significant shoulder dystocia, only shoulder skin-to-skin distance was significantly greater and OC was significantly lower in comparison with the control group, whereas in those who had significant shoulder dystocia, only SOD and MTD were significantly lower in comparison with the control group.

CONCLUSIONS

MRI-derived fetal size, fetal shoulder measurements and maternal pelvimetry are associated with shoulder dystocia. Future studies could incorporate these measurements into a reliable predictive model for shoulder dystocia. © 2025 International Society of Ultrasound in Obstetrics and Gynecology.

Novel machine learning applications in peripartum care: a scoping review

OBJECTIVE

Machine learning (ML), a subtype of artificial intelligence (AI), presents predictive modeling and dynamic diagnostic tools to facilitate early interventions and improve decision-making. Considering the global challenges of maternal, fetal, and neonatal morbidity and mortality, ML holds the potential to enable significant improvements in maternal and neonatal health outcomes. We aimed to conduct a comprehensive review of ML applications in peripartum care, summarizing the potential of these tools to enhance clinical decision-making and identifying emerging trends and research gaps.

DATA SOURCES

We conducted a scoping review on MEDLINE, Cochrane Library, and EMBASE databases from inception to April 2024. We gathered additional relevant studies through snowball sampling. We meticulously screened titles and abstracts and chose full-text articles for further analysis.

STUDY ELIGIBILITY CRITERIA

We included primary research articles and abstracts focusing on pregnant individuals, employing ML methods for peripartum care.

STUDY APPRAISAL AND SYNTHESIS METHODS

No formal quality assessment was performed. Data were extracted using a custom template to capture study characteristics and ML models. Findings were synthesized using summary tables and figures to highlight key trends and results.

RESULTS

Among 406 studies, 78% were published within the last five years. Most studies originated from high-income or well-resourced countries, with 27% from North America (including 24% from the United States) and 34% from Asia, predominantly China (18%). Studies from low- and middle-income regions were notably scarce, reflecting significant regional disparities. Predictive modeling tasks were the most prevalent (59%), followed by classification tasks (29%). Supervised learning dominated (90%), with algorithms such as Support Vector Machines, Random Forests, and Logistic Regression most commonly used. Key topics included fetal distress and acidemia (32%), preterm birth (22%), mode of delivery (13%), and birth weight (13%). Notably, Explainable AI methods were utilized in only 19% of studies, and external validation was performed in just 5%. Despite these advancements, only 1% of models resulted in accessible clinical tools, and none were fully integrated into healthcare systems.

CONCLUSIONS

ML holds significant potential to enhance peripartum care by improving diagnostic accuracy and predictive capabilities. However, realizing this potential requires responsible AI practices, including robust validation with external datasets, prospective investigations across diverse populations, and the development of digital and data infrastructure for seamless integration into electronic health records. Additionally, transparent AI that provides insights into risk stratification logic is essential for clinician trust in ML tools. Future research should address understudied areas, prioritize neglected low-income settings, and explore advanced ML approaches to improve maternal and neonatal outcomes.

Construction of an explanatory model for predicting hepatotoxicity: a case study of the potentially hepatotoxic components of Gardenia jasminoides

It is well-known that the hepatotoxicity of drugs can significantly influence their clinical use. Despite their effective therapeutic efficacy, many drugs are severely limited in clinical applications due to significant hepatotoxicity. In response, researchers have created several machine learning-based hepatotoxicity prediction models for use in drug discovery and development. Researchers aim to predict the potential hepatotoxicity of drugs to enhance their utility. However, current hepatotoxicity prediction models often suffer from being unverified, and they fail to capture the detailed toxicological structures of predicted hepatotoxic compounds. Using the 56 chemical constituents of Gardenia jasminoides as examples, we validated the trained hepatotoxicity prediction model through literature reviews, principal component analysis (PCA), and structural comparison methods. Ultimately, we successfully developed a model with strong predictive performance and conducted visual validation. Interestingly, we discovered that the predicted hepatotoxic chemical constituents of Gardenia possess both toxic and therapeutic effects, which are likely dose-dependent. This discovery greatly contributes to our understanding of the dual nature of drug-induced hepatotoxicity.

Molecular classification of geriatric breast cancer displays distinct senescent subgroups of prognostic significance

Breast cancer in the elderly presents distinct biological characteristics and clinical treatment responses compared with cancer in younger patients. Comprehensive Geriatric Assessment is recommended for evaluating treatment efficacy in elderly cancer patients based on physiological classification. However, research on molecular classification in older cancer patients remains insufficient. In this study, we identified two subgroups with distinct senescent clusters among geriatric breast cancer patients through multi-omics analysis. Using various machine learning algorithms, we developed a comprehensive scoring model called "Sene_Signature," which more accurately distinguished elderly breast cancer patients compared with existing methods and better predicted their prognosis. The Sene_Signature was correlated with tumor immune cell infiltration, as supported by single-cell transcriptomics, RNA sequencing, and pathological data. Furthermore, we observed increased drug responsiveness in patients with a high Sene_Signature to treatments targeting the epidermal growth factor receptor and cell-cycle pathways. We also established a user-friendly web platform to assist investigators in assessing Sene_Signature scores and predicting treatment responses for elderly breast cancer patients. In conclusion, we developed a novel model for evaluating prognosis and therapeutic responses, providing a potential molecular classification that assists in the pre-treatment assessment of geriatric breast cancer.

Prediction of gestational diabetes mellitus at the first trimester: machine-learning algorithms

PURPOSE

Short- and long-term complications of gestational diabetes mellitus (GDM) involving pregnancies and offspring warrant the development of an effective individualized risk prediction model to reduce and prevent GDM together with its associated co-morbidities. The aim is to use machine learning (ML) algorithms to study data gathered throughout the first trimester in order to predict GDM.

METHODS

Two independent cohorts with forty-five features gathered through first trimester were included. We constructed prediction models based on three different algorithms and traditional logistic regression, and deployed additional two ensemble algorithms to identify the importance of individual features.

RESULTS

4799 and 2795 pregnancies were included in the Xinhua Hospital Chongming branch (XHCM) and the Shanghai Pudong New Area People's Hospital (SPNPH) cohorts, respectively. Extreme gradient boosting (XGBoost) predicted GDM with moderate performance (the area under the receiver operating curve (AUC) = 0.75) at pregnancy initiation and good-to-excellent performance (AUC = 0.99) at the end of the first trimester in the XHCM cohort. The trained XGBoost showed moderate performance in the SPNPH cohort (AUC = 0.83). The top predictive features for GDM diagnosis were pre-pregnancy BMI and maternal abdominal circumference at pregnancy initiation, and FPG and HbA1c at the end of the first trimester.

CONCLUSION

Our work demonstrated that ML models based on the data gathered throughout the first trimester achieved moderate performance in the external validation cohort.

Development and validation of a deep-learning model to predict 10-year atherosclerotic cardiovascular disease risk from retinal images using the UK Biobank and EyePACS 10K datasets

Background

Atherosclerotic cardiovascular disease (ASCVD) is a leading cause of death globally, and early detection of high-risk individuals is essential for initiating timely interventions. The authors aimed to develop and validate a deep learning (DL) model to predict an individual's elevated 10-year ASCVD risk score based on retinal images and limited demographic data.

Methods

The study used 89,894 retinal fundus images from 44,176 UK Biobank participants (96% non-Hispanic White, 5% diabetic) to train and test the DL model. The DL model was developed using retinal images plus age, race/ethnicity, and sex at birth to predict an individual's 10-year ASCVD risk score using the pooled cohort equation (PCE) as the ground truth. This model was then tested on the US EyePACS 10K dataset (5.8% non-Hispanic White, 99.9% diabetic), composed of 18,900 images from 8969 diabetic individuals. Elevated ASCVD risk was defined as a PCE score of ≥7.5%.

Results

In the UK Biobank internal validation dataset, the DL model achieved an area under the receiver operating characteristic curve of 0.89, sensitivity 84%, and specificity 90%, for detecting individuals with elevated ASCVD risk scores. In the EyePACS 10K and with the addition of a regression-derived diabetes modifier, it achieved sensitivity 94%, specificity 72%, mean error -0.2%, and mean absolute error 3.1%.

Conclusion

This study demonstrates that DL models using retinal images can provide an additional approach to estimating ASCVD risk, as well as the value of applying DL models to different external datasets and opportunities about ASCVD risk assessment in patients living with diabetes.

Shoulder dystocia in deliveries of neonates <3500 grams

OBJECTIVES

To study risk factors for shoulder dystocia (ShD) among women delivering <3500 g newborn.

METHODS

A retrospective case-control study of all term live-singleton deliveries during 2011-2019. Women with neonatal birthweight <3500 g were included. We compared cases of ShD to other deliveries by univariate and multivariable regression.

RESULTS

There were 79/41 092 (0.19%) cases of ShD among neonates <3500 g. In multivariable regression analysis, the following factors were independently associated with ShD; operative vaginal delivery (odds ratio [OR] 2.78; 95% confidence interval [CI]: 1.28-6.02, P = 0.009), vaginal birth after cesarean (VBAC, OR 2.74; 1.22-6.13, P = 0.010), sonographic abdominal circumference to biparietal diameter ratio (3.73 among ShD vs. 3.62, OR 1.35; 95% CI: 1.12-1.63, P = 0.001) and sonographic abdominal circumference to head circumference ratio (1.036 among ShD vs. 1.011, OR 3.04; 95% CI: 1.006-9.23, P = 0.049).

CONCLUSIONS

There is an association between operative vaginal delivery and ShD also in deliveries <3500 g. Importantly, the proportions between the fetal head and abdominal circumference are a better predictor of ShD than the newborn fetal weight and VBAC is associated with ShD.

Interpretable Predictive Models to Understand Risk Factors for Maternal and Fetal Outcomes

Although most pregnancies result in a good outcome, complications are not uncommon and can be associated with serious implications for mothers and babies. Predictive modeling has the potential to improve outcomes through a better understanding of risk factors, heightened surveillance for high-risk patients, and more timely and appropriate interventions, thereby helping obstetricians deliver better care. We identify and study the most important risk factors for four types of pregnancy complications: (i) severe maternal morbidity, (ii) shoulder dystocia, (iii) preterm preeclampsia, and (iv) antepartum stillbirth. We use an Explainable Boosting Machine (EBM), a high-accuracy glass-box learning method, for the prediction and identification of important risk factors. We undertake external validation and perform an extensive robustness analysis of the EBM models. EBMs match the accuracy of other black-box ML methods, such as deep neural networks and random forests, and outperform logistic regression, while being more interpretable. EBMs prove to be robust. The interpretability of the EBM models reveal surprising insights into the features contributing to risk (e.g., maternal height is the second most important feature for shoulder dystocia) and may have potential for clinical application in the prediction and prevention of serious complications in pregnancy.

Artificial intelligence as a new answer to old challenges in maternal-fetal medicine and obstetrics

BACKGROUND

Following the latest trends in the development of artificial intelligence (AI), the possibility of processing an immense amount of data has created a breakthrough in the medical field. Practitioners can now utilize AI tools to advance diagnostic protocols and improve patient care.

OBJECTIVE

The aim of this article is to present the importance and modalities of AI in maternal-fetal medicine and obstetrics and its usefulness in daily clinical work and decision-making process.

METHODS

A comprehensive literature review was performed by searching PubMed for articles published from inception up until August 2023, including the search terms "artificial intelligence in obstetrics", "maternal-fetal medicine", and "machine learning" combined through Boolean operators. In addition, references lists of identified articles were further reviewed for inclusion.

RESULTS

According to recent research, AI has demonstrated remarkable potential in improving the accuracy and timeliness of diagnoses in maternal-fetal medicine and obstetrics, e.g., advancing perinatal ultrasound technique, monitoring fetal heart rate during labor, or predicting mode of delivery. The combination of AI and obstetric ultrasound can help optimize fetal ultrasound assessment by reducing examination time and improving diagnostic accuracy while reducing physician workload.

CONCLUSION

The integration of AI in maternal-fetal medicine and obstetrics has the potential to significantly improve patient outcomes, enhance healthcare efficiency, and individualized care plans. As technology evolves, AI algorithms are likely to become even more sophisticated. However, the successful implementation of AI in maternal-fetal medicine and obstetrics needs to address challenges related to interpretability and reliability.

An evaluation of cervical maturity for Chinese women with labor induction by machine learning and ultrasound images

BACKGROUND

To evaluate the improvement of evaluation accuracy of cervical maturity for Chinese women with labor induction by adding objective ultrasound data and machine learning models to the existing traditional Bishop method.

METHODS

The machine learning model was trained and tested using 101 sets of data from pregnant women who were examined and had their delivery in Peking University Third Hospital in between December 2019 and January 2021. The inputs of the model included cervical length, Bishop score, angle, age, induced labor time, measurement time (MT), measurement time to induced labor time (MTILT), method of induced labor, and primiparity/multiparity. The output of the model is the predicted time from induced labor to labor. Our experiments analyzed the effectiveness of three machine learning models: XGBoost, CatBoost and RF(Random forest). we consider the root-mean-squared error (RMSE) and the mean absolute error (MAE) as the criterion to evaluate the accuracy of the model. Difference was compared using t-test on RMSE between the machine learning model and the traditional Bishop score.

RESULTS

The mean absolute error of the prediction result of Bishop scoring method was 19.45 h, and the RMSE was 24.56 h. The prediction error of machine learning model was lower than the Bishop score method. Among the three machine learning models, the MAE of the model with the best prediction effect was 13.49 h and the RMSE was 16.98 h. After selection of feature the prediction accuracy of the XGBoost and RF was slightly improved. After feature selection and artificially removing the Bishop score, the prediction accuracy of the three models decreased slightly. The best model was XGBoost (p = 0.0017). The p-value of the other two models was < 0.01.

CONCLUSION

In the evaluation of cervical maturity, the results of machine learning method are more objective and significantly accurate compared with the traditional Bishop scoring method. The machine learning method is a better predictor of cervical maturity than the traditional Bishop method.

Early Prediction Model of Macrosomia Using Machine Learning for Clinical Decision Support

The condition of fetal overgrowth, also known as macrosomia, can cause serious health complications for both the mother and the infant. It is crucial to identify high-risk macrosomia-relevant pregnancies and intervene appropriately. Despite this need, there are several gaps in research related to macrosomia, including limited predictive models, insufficient machine learning applications, ineffective interventions, and inadequate understanding of how to integrate machine learning models into clinical decision-making. To address these gaps, we developed a machine learning-based model that uses maternal characteristics and medical history to predict macrosomia. Three different algorithms, namely logistic regression, support vector machine, and random forest, were used to develop the model. Based on the evaluation metrics, the logistic regression algorithm provided the best results among the three. The logistic regression algorithm was chosen as the final algorithm to predict macrosomia. The hyper parameters of the logistic regression model were tuned using cross-validation to achieve the best possible performance. Our results indicate that machine learning-based models have the potential to improve macrosomia prediction and enable appropriate interventions for high-risk pregnancies, leading to better health outcomes for both mother and fetus. By leveraging machine learning algorithms and addressing research gaps related to macrosomia, we can potentially reduce the health risks associated with this condition and make informed decisions about high-risk pregnancies.

Prediction of birthweight and risk of macrosomia in pregnancies complicated by diabetes

BACKGROUND

Antenatal detection of accelerated fetal growth and macrosomia in pregnancies complicated by diabetes mellitus is important for patient counseling and management. Sonographic fetal weight estimation is the most commonly used tool to predict birthweight and macrosomia. However, the predictive accuracy of sonographic fetal weight estimation for these outcomes is limited. In addition, an up-to-date sonographic fetal weight estimation is often unavailable before birth. This may result in a failure to identify macrosomia, especially in pregnancies complicated by diabetes mellitus where care providers might underestimate fetal growth rate. Therefore, there is a need for better tools to detect and alert care providers to the potential risk of accelerated fetal growth and macrosomia.

OBJECTIVE

This study aimed to develop and validate prediction models for birthweight and macrosomia in pregnancies complicated by diabetes mellitus.

STUDY DESIGN

This was a completed retrospective cohort study of all patients with a singleton live birth at ≥36 weeks of gestation complicated by preexisting or gestational diabetes mellitus observed at a single tertiary center between January 2011 and May 2022. Candidate predictors included maternal age, parity, type of diabetes mellitus, information from the most recent sonographic fetal weight estimation (including estimated fetal weight, abdominal circumference z score, head circumference-to-abdomen circumference z score ratio, and amniotic fluid), fetal sex, and the interval between ultrasound examination and birth. The study outcomes were macrosomia (defined as birthweights >4000 and >4500 g), large for gestational age (defined as a birthweight >90th percentile for gestational age), and birthweight (in grams). Multivariable logistic regression models were used to estimate the probability of dichotomous outcomes, and multivariable linear regression models were used to estimate birthweight. Model discrimination and predictive accuracy were calculated. Internal validation was performed using the bootstrap resampling technique.

RESULTS

A total of 2465 patients met the study criteria. Most patients had gestational diabetes mellitus (90%), 6% of patients had type 2 diabetes mellitus, and 4% of patients had type 1 diabetes mellitus. The overall proportions of infants with birthweights >4000 g, >4500 g, and >90th percentile for gestational age were 8%, 1%, and 12%, respectively. The most contributory predictor variables were estimated fetal weight, abdominal circumference z score, ultrasound examination to birth interval, and type of diabetes mellitus. The models for the 3 dichotomous outcomes had high discriminative accuracy (area under the curve receiver operating characteristic curve, 0.929-0.979), which was higher than that achieved with estimated fetal weight alone (area under the curve receiver operating characteristic curve, 0.880-0.931). The predictive accuracy of the models had high sensitivity (87%-100%), specificity (84%-92%), and negative predictive values (84%-92%). The predictive accuracy of the model for birthweight had low systematic and random errors (0.6% and 7.5%, respectively), which were considerably smaller than the corresponding errors achieved with estimated fetal weight alone (-5.9% and 10.8%, respectively). The proportions of estimates within 5%, 10%, and 15% of the actual birthweight were high (52.3%, 82.9%, and 94.9%, respectively).

CONCLUSION

The prediction models developed in the current study were associated with greater predictive accuracy for macrosomia, large for gestational age, and birthweight than the current standard of care that includes estimated fetal weight alone. These models may assist care providers in counseling patients regarding the optimal timing and mode of delivery.

Machine learning and disease prediction in obstetrics

Machine learning technologies and translation of artificial intelligence tools to enhance the patient experience are changing obstetric and maternity care. An increasing number of predictive tools have been developed with data sourced from electronic health records, diagnostic imaging and digital devices. In this review, we explore the latest tools of machine learning, the algorithms to establish prediction models and the challenges to assess fetal well-being, predict and diagnose obstetric diseases such as gestational diabetes, pre-eclampsia, preterm birth and fetal growth restriction. We discuss the rapid growth of machine learning approaches and intelligent tools for automated diagnostic imaging of fetal anomalies and to asses fetoplacental and cervix function using ultrasound and magnetic resonance imaging. In prenatal diagnosis, we discuss intelligent tools for magnetic resonance imaging sequencing of the fetus, placenta and cervix to reduce the risk of preterm birth. Finally, the use of machine learning to improve safety standards in intrapartum care and early detection of complications will be discussed. The demand for technologies to enhance diagnosis and treatment in obstetrics and maternity should improve frameworks for patient safety and enhance clinical practice.

Utilizing machine learning to predict unplanned cesarean delivery

OBJECTIVE

To develop a comprehensive machine learning (ML) model predicting unplanned cesarean delivery (uCD) among singleton pregnancies based on features available at admission to labor.

METHODS

A retrospective cohort study from a tertiary medical center. Women with singleton vertex pregnancy of 34 weeks or more admitted for vaginal delivery between March 2011 and May 2019 were included. The cohort was divided into training (80%) and validation (20%) data sets. A separate cohort between June 2019 and April 2021 served as a test data set. Features selection was performed using a Random Forest ML algorithm.

RESULTS

The study population included 73 667 women, of which 4125 (6.33%) underwent uCD. The final model consisted of 13 features, based on prediction importance. The XGBoost model performed best with areas under the curve for the training, validation, and test data sets of 0.874, 0.839, and 0.840, respectively. The model showed a 65% positive predictive value for uCD among women in the 100th centile group, and a 99% or more negative predictive value in the less than 50th centile group. Positive and negative predictive values remained high among subgroups with high pretest probability of uCD.

CONCLUSION

An ML model for the prediction of uCD provides clinically useful risk stratification that remains accurate across gestational weeks 34-42 and among clinical risk groups. The model may be clinically useful for physicians and women admitted for labor.

SYNOPSIS

A machine learning model predicts unplanned cesarean delivery and can inform women's individualized decision making.

Use of artificial intelligence in obstetrics: not quite ready for prime time

Artificial intelligence is finding several applications in healthcare settings. This study aimed to report evidence on the effectiveness of artificial intelligence application in obstetrics. Through a narrative review of literature, we described artificial intelligence use in different obstetrical areas as follows: prenatal diagnosis, fetal heart monitoring, prediction and management of pregnancy-related complications (preeclampsia, preterm birth, gestational diabetes mellitus, and placenta accreta spectrum), and labor. Artificial intelligence seems to be a promising tool to help clinicians in daily clinical activity. The main advantages that emerged from this review are related to the reduction of inter- and intraoperator variability, time reduction of procedures, and improvement of overall diagnostic performance. However, nowadays, the diffusion of these systems in routine clinical practice raises several issues. Reported evidence is still very limited, and further studies are needed to confirm the clinical applicability of artificial intelligence. Moreover, better training of clinicians designed to use these systems should be ensured, and evidence-based guidelines regarding this topic should be produced to enhance the strengths of artificial systems and minimize their limits.

Current developments in artificial intelligence from obstetrics and gynecology to urogynecology

In today's medical practice clinicians need to struggle with a huge amount of data to improve the outcomes of the patients. Sometimes one clinician needs to deal with thousands of ultrasound images or hundred papers of laboratory results. To overcome this shortage, computers get in help of human beings and they are educated under the term "artificial intelligence." We were using artificial intelligence in our daily lives (i.e., Google, Netflix, etc.), but applications in medicine are relatively new. In obstetrics and gynecology, artificial intelligence models mostly use ultrasound images for diagnostic purposes but nowadays researchers started to use other medical recordings like non-stress tests or urodynamics study results to develop artificial intelligence applications. Urogynecology is a developing subspecialty of obstetrics and gynecology, and articles about artificial intelligence in urogynecology are limited but in this review, we aimed to increase clinicians' knowledge about this new approach.

Antenatal risk score for prediction of shoulder dystocia with focus on fetal ultrasound data

BACKGROUND

Shoulder dystocia is one of the most threatening complications during delivery, and although it is difficult to predict, individual risk should be considered when counseling for mode of delivery.

OBJECTIVE

This study aimed to develop and validate a risk score for shoulder dystocia based on fetal ultrasound and maternal data from 15,000 deliveries.

STUDY DESIGN

Data were retrospectively obtained of deliveries in 3 tertiary centers between 2014 and 2017 for the derivation cohort and between 2018 and 2020 for the validation cohort. Inclusion criteria were singleton pregnancy, vaginal delivery in cephalic presentation at ≥37+0 weeks' gestation, and fetal biometry data available within 2 weeks of delivery. Independent predictors were determined by multivariate regression analysis in the derivation cohort, and a score was developed on the basis of the effect of the predictors.

RESULTS

The derivation cohort consisted of 7396 deliveries with a 0.91% rate of shoulder dystocia, and the validation cohort of 7965 deliveries with a 1.0% rate of shoulder dystocia. Among all women, 13.8% had diabetes mellitus, and 12.1% were obese (body mass index ≥30 kg/m2). Independent risk factors in the derivation cohort were: estimated fetal weight ≥4250 g (odds ratio, 4.27; P=.002), abdominal-head-circumference ≥2.5 cm (odds ratio, 3.96; P<.001), and diabetes mellitus (odds ratio, 2.18; P=.009). On the basis of the strength of effect, a risk score was developed: estimated fetal weight ≥4250 g=2, abdominal-head-circumference ≥2.5 cm=2, and diabetes mellitus=1. The risk score predicted shoulder dystocia with moderate discriminatory ability (area under the receiver-operating characteristic curve, 0.69; P<.001; area under the receiver-operating characteristic curve, 0.71; P<.001) and good calibration (Hosmer-Lemeshow goodness-of-fit; P=.466; P=.167) for the derivation and validation cohorts, respectively. With 1 score point, 16 shoulder dystocia cases occurred in 1764 deliveries, with 0.6% shoulder dystocia incidence and a number needed to treat with cesarean delivery to avoid 1 case of shoulder dystocia of 172 (2 points: 38/1809, 2.1%, 48; 3 points: 18/336, 5.4%, 19; 4 points: 10/96, 10.5%, 10; and 5 points: 5/20, 25%, 4); 40.8% of the shoulder dystocia cases occurred without risk factors.

CONCLUSION

The presented risk score for shoulder dystocia may act as a supplemental tool for the clinical decision-making regarding mode of delivery. According to our score model, in pregnancies with a score ≤2, meaning having solely estimated fetal weight ≥4250 g, or abdominal-head-circumference ≥2.5, or diabetes mellitus, cesarean delivery for prevention of shoulder dystocia should not be recommended because of the high number needed to treat to avoid 1 case of shoulder dystocia. Conversely, in patients with a score of ≥4 with or without diabetes mellitus, cesarean delivery may be considered. However, in 40% of the shoulder dystocia cases, no risk factors had been present.

The association between an oral glucose tolerance test performed at term pregnancy and obstetric outcomes

Aims

Assessing the value of oral glucose tolerance test performed at term pregnancy in identifying obstetric complications.

Methods

Retrospective cohort study of women with a normal 50 g glucose challenge test who also had an oral glucose tolerance test at term (defined as at or after 37 weeks of gestation). Comparison between the pathological and normal oral glucose tolerance test groups was performed.

Results

The mean glucose in the glucose challenge test of women in the normal oral glucose tolerance test (n = 256) group was lower than that in the pathological oral glucose tolerance test (N = 16) group (105 ± 17 mg/dl (5.8 ± 0.9 mmol/l) vs 117 ± 13 mg/dl (6.5 ± 0.7 mmol/l), p = 0.007). Relevant obstetrical complications did not differ significantly between the groups. Of note, in the pathological oral glucose tolerance test group only one woman delivered a macrosomic infant.

Conclusions

A pathological oral glucose tolerance test performed at term was unable to identify women at risk for impaired glucose metabolism-related obstetric complications and is therefore of limited clinical value and seems to be unjustified.

Machine learning is an effective method to predict the 90-day prognosis of patients with transient ischemic attack and minor stroke

OBJECTIVE

We aimed to investigate factors related to the 90-day poor prognosis (mRS≥3) in patients with transient ischemic attack (TIA) or minor stroke, construct 90-day poor prognosis prediction models for patients with TIA or minor stroke, and compare the predictive performance of machine learning models and Logistic model.

METHOD

We selected TIA and minor stroke patients from a prospective registry study (CNSR-III). Demographic characteristics,smoking history, drinking history(≥20g/day), physiological data, medical history,secondary prevention treatment, in-hospital evaluation and education,laboratory data, neurological severity, mRS score and TOAST classification of patients were assessed. Univariate and multivariate logistic regression analyses were performed in the training set to identify predictors associated with poor outcome (mRS≥3). The predictors were used to establish machine learning models and the traditional Logistic model, which were randomly divided into the training set and test set according to the ratio of 70:30. The training set was used to construct the prediction model, and the test set was used to evaluate the effect of the model. The evaluation indicators of the model included the area under the curve (AUC) of the discrimination index and the Brier score (or calibration plot) of the calibration index.

RESULT

A total of 10967 patients with TIA and minor stroke were enrolled in this study, with an average age of 61.77 ± 11.18 years, and women accounted for 30.68%. Factors associated with the poor prognosis in TIA and minor stroke patients included sex, age, stroke history, heart rate, D-dimer, creatinine, TOAST classification, admission mRS, discharge mRS, and discharge NIHSS score. All models, both those constructed by Logistic regression and those by machine learning, performed well in predicting the 90-day poor prognosis (AUC >0.800). The best performing AUC in the test set was the Catboost model (AUC=0.839), followed by the XGBoost, GBDT, random forest and Adaboost model (AUCs equal to 0.838, 0, 835, 0.832, 0.823, respectively). The performance of Catboost and XGBoost in predicting poor prognosis at 90-day was better than the Logistic model, and the difference was statistically significant(P<0.05). All models, both those constructed by Logistic regression and those by machine learning had good calibration.

CONCLUSION

Machine learning algorithms were not inferior to the Logistic regression model in predicting the poor prognosis of patients with TIA and minor stroke at 90-day. Among them, the Catboost model had the best predictive performance. All models provided good discrimination.

Accuracy of Fetal Biacromial Diameter and Derived Ultrasonographic Parameters to Predict Shoulder Dystocia: A Prospective Observational Study

Background and Objectives: Shoulder dystocia (ShD) is one of most dangerous obstetric complication. The objective of this study was to determine if the ultrasonographic fetal biacromial diameter (BA) and derived parameters could predict ShD in uncomplicated term pregnancies. Materials and Methods: We conducted a prospective observational study in a tertiary care university hospital from March 2021 to February 2022. We included all full-term pregnancies accepted for delivery that received an accurate ultrasonography (USG) scan before delivery. USG biometry and estimated fetal weight (EFW) were collected. Therefore, we evaluated the diameter of the mid-arm, the transverse thoracic diameter (TTD) and the biacromial diameter (BA). BA was estimated using Youssef’s formula: TTD + 2 mid-arm diameters. The primary outcome was the evaluation of BA and its related parameters (BA/biparietal diameter (BPD), BA/head circumference (HC) and BA–BPD in fetuses with ShD versus fetuses without ShD. Diagnostic accuracy for ShD of BA, BA/BPD, BA/HC and BA–BPD was evaluated using receiver operator curve (ROC) analysis. Results: 90 women were included in the analysis, four of these had ShD and required extra maneuvers after head delivery. BA was increased in fetuses with ShD (150.4 cm; 95% CI 133.2 cm to 167.6 cm) compared to no-ShD (133.5 cm; 95% CI 130.1 cm to 137.0 cm; p = 0.04). Significant differences were also found between ShD and no-ShD groups for BA/BPD (1.66 (95% CI 1.46 to 1.86) vs. 1.44 (95% CI 1.41 to 1.48); p = 0.04), BA/HC (0.45 (95% CI 0.40 to 0.49) vs. 0.39 (95% CI 0.38 to 0.40); p = 0.01), BA–BPD (60.0 mm (95% CI 42.4 to 77.6 cm) vs. 41.4 (95% CI 38.2 to 44.6); p = 0.03), respectively. ROC analysis showed an overall good accuracy for ShD, with an AUC of 0.821 (p = 0.001) for BA alone and 0.881 (p = 0.001), 0.857 (p = 0.016) and 0.867 (p = 0.013) for BA/BPD, BA–BPD and BA/HC, respectively. Conclusions: BA alone, as well as BA/BPD, BA/HC and BA–BPD might be useful predictors of ShD in uncomplicated term pregnancies. However, such evidence needs extensive confirmation by means of additional studies with large sample sizes, especially in case of pregnancies at high risk for ShD (i.e., gestational diabetes).

Prediction of Emergency Cesarean Section Using Machine Learning Methods: Development and External Validation of a Nationwide Multicenter Dataset in Republic of Korea

This study was a multicenter retrospective cohort study of term nulliparous women who underwent labor, and was conducted to develop an automated machine learning model for prediction of emergent cesarean section (CS) before onset of labor. Nine machine learning methods of logistic regression, random forest, Support Vector Machine (SVM), gradient boosting, extreme gradient boosting (XGBoost), light gradient boosting machine (LGBM), k-nearest neighbors (KNN), Voting, and Stacking were applied and compared for prediction of emergent CS during active labor. External validation was performed using a nationwide multicenter dataset for Korean fetal growth. A total of 6549 term nulliparous women was included in the analysis, and the emergent CS rate was 16.1%. The C-statistics values for KNN, Voting, XGBoost, Stacking, gradient boosting, random forest, LGBM, logistic regression, and SVM were 0.6, 0.69, 0.64, 0.59, 0.66, 0.68, 0.68, 0.7, and 0.69, respectively. The logistic regression model showed the best predictive performance with an accuracy of 0.78. The machine learning model identified nine significant variables of maternal age, height, weight at pre-pregnancy, pregnancy-associated hypertension, gestational age, and fetal sonographic findings. The C-statistic value for the logistic regression machine learning model in the external validation set (1391 term nulliparous women) was 0.69, with an overall accuracy of 0.68, a specificity of 0.83, and a sensitivity of 0.41. Machine learning algorithms with clinical and sonographic parameters at near term could be useful tools to predict individual risk of emergent CS during active labor in nulliparous women.

Novel artificial intelligence approach for automatic differentiation of fetal occiput anterior and non-occiput anterior positions during labor

OBJECTIVES

To describe a newly developed machine-learning (ML) algorithm for the automatic recognition of fetal head position using transperineal ultrasound (TPU) during the second stage of labor and to describe its performance in differentiating between occiput anterior (OA) and non-OA positions.

METHODS

This was a prospective cohort study including singleton term (> 37 weeks of gestation) pregnancies in the second stage of labor, with a non-anomalous fetus in cephalic presentation. Transabdominal ultrasound was performed to determine whether the fetal head position was OA or non-OA. For each case, one sonographic image of the fetal head was then acquired in an axial plane using TPU and saved for later offline analysis. Using the transabdominal sonographic diagnosis as the gold standard, a ML algorithm based on a pattern-recognition feed-forward neural network was trained on the TPU images to discriminate between OA and non-OA positions. In the training phase, the model tuned its parameters to approximate the training data (i.e. the training dataset) such that it would identify correctly the fetal head position, by exploiting geometric, morphological and intensity-based features of the images. In the testing phase, the algorithm was blinded to the occiput position as determined by transabdominal ultrasound. Using the test dataset, the ability of the ML algorithm to differentiate OA from non-OA fetal positions was assessed in terms of diagnostic accuracy. The F1 -score and precision-recall area under the curve (PR-AUC) were calculated to assess the algorithm's performance. Cohen's kappa (κ) was calculated to evaluate the agreement between the algorithm and the gold standard.

RESULTS

Over a period of 24 months (February 2018 to January 2020), at 15 maternity hospitals affiliated to the International Study group on Labor ANd Delivery Sonography (ISLANDS), we enrolled into the study 1219 women in the second stage of labor. On the basis of transabdominal ultrasound, they were classified as OA (n = 801 (65.7%)) or non-OA (n = 418 (34.3%)). From the entire cohort (OA and non-OA), approximately 70% (n = 824) of the patients were assigned randomly to the training dataset and the rest (n = 395) were used as the test dataset. The ML-based algorithm correctly classified the fetal occiput position in 90.4% (357/395) of the test dataset, including 224/246 with OA (91.1%) and 133/149 with non-OA (89.3%) fetal head position. Evaluation of the algorithm's performance gave an F1 -score of 88.7% and a PR-AUC of 85.4%. The algorithm showed a balanced performance in the recognition of both OA and non-OA positions. The robustness of the algorithm was confirmed by high agreement with the gold standard (κ = 0.81; P < 0.0001).

CONCLUSIONS

This newly developed ML-based algorithm for the automatic assessment of fetal head position using TPU can differentiate accurately, in most cases, between OA and non-OA positions in the second stage of labor. This algorithm has the potential to support not only obstetricians but also midwives and accoucheurs in the clinical use of TPU to determine fetal occiput position in the labor ward. © 2021 International Society of Ultrasound in Obstetrics and Gynecology.

Artificial intelligence in obstetrics

This study reviews recent advances on the application of artificial intelligence for the early diagnosis of various maternal-fetal conditions such as preterm birth and abnormal fetal growth. It is found in this study that various machine learning methods have been successfully employed for different kinds of data capture with regard to early diagnosis of maternal-fetal conditions. With the more popular use of artificial intelligence, ethical issues should also be considered accordingly.

Antenatal risk prediction of shoulder dystocia: influence of diabetes and obesity: a multicenter study

PURPOSE

To estimate the risk of shoulder dystocia (SD) in pregnancies with/without maternal diabetes or obesity; to identify antenatal maternal and fetal ultrasound-derived risk factors and calculate their contributions.

METHODS

A multicenter retrospective analysis of 13,428 deliveries in three tertiary hospitals (2014-2017) with fetal ultrasound data ≤ 14 days prior to delivery (n = 7396).

INCLUSION CRITERIA

singleton pregnancies in women ≥ 18 years old; vertex presentation; vaginal delivery at ≥ 37 weeks of gestation. Estimated fetal weight (EFW) and birth weight (BW) were categorized by steps of 250 g. To evaluate risk factors, a model was performed using ultrasound data with SD as the dependent variable.

RESULTS

Diabetes was present in 9.3%; BMI ≥ 30 kg/m² in 10.4% and excessive weight gain in 39.8%. The total SD rate was 0.9%, with diabetes 2.0% and with obesity 1.9%. These increased with BW 4250-4499 g compared to 4000-4249 g in women with diabetes (12.1% vs 1.9%, P = 0.010) and without (6.1% vs 1.6%, P < 0.001) and at the same BW threshold for women with obesity (9.6% vs 0.6%, P = 0.002) or without (6.4% vs 1.8%, P < 0.001). Rates increased similarly for EFW at 4250 g and for AC-HC at 2.5 cm. Independent risk factors for SD were EFW ≥ 4250 g (OR 3.8, 95% CI 1.5-9.4), AC-HC ≥ 2.5 cm (OR 3.1, 95% CI 1.3-7.5) and diabetes (OR 2.2, 95% CI 1.2-4.0). HC/AC ratio, obesity, excessive weight gain and labor induction were not significant.

CONCLUSION

Independent of diabetes, which remains a risk factor for SD, a significant increase may be expected if the EFW is ≥ 4250 g and AC-HC is ≥ 2.5 cm.

Shoulder dystocia and composite adverse outcomes for the maternal-neonatal dyad

BACKGROUND

Although the neonatal morbidity associated with shoulder dystocia are well known, the maternal morbidity caused by this obstetrical emergency is infrequently reported.

OBJECTIVE

This study aimed to assess the composite adverse maternal and neonatal outcomes among vaginal deliveries (at 34 weeks or later) with and without shoulder dystocia.

STUDY DESIGN

This is a secondary analysis of the Consortium of Safe Labor, an observational obstetrical cohort of all vaginal deliveries occurring at 19 hospitals (from 2002-2008) and for which data on the occurrence of shoulder dystocia were available. The composite adverse maternal outcome included third- or fourth-degree perineal laceration, postpartum hemorrhage (>500 cc blood loss for a vaginal delivery and >1000 cc blood loss for cesarean delivery), blood transfusion, chorioamnionitis, endometritis, thromboembolism, admission to intensive care unit, or maternal death. The composite adverse neonatal outcome included an Apgar score of <7 at 5 minutes, a birth injury, neonatal seizure, hypoxic ischemic encephalopathy, or neonatal death. A multivariable Poisson regression was used to estimate the adjusted relative risks with 95% confidence intervals. The area under the receiver operating characteristic curve was constructed to determine if clinical factors would identify shoulder dystocia.

RESULTS

Of the 228,438 women in the overall cohort, 130,008 (59.6%) met the inclusion criteria, and among them, shoulder dystocia was documented in 2159 (1.7%) cases. The rate of composite maternal morbidity was significantly higher among deliveries with shoulder dystocia (14.7%) than without (8.6%; adjusted relative risk, 1.71; 95% confidence interval, 1.64-2.01). The most common maternal morbidity with shoulder dystocia was a third- or fourth-degree laceration (adjusted relative risk, 2.82; 95% confidence interval, 2.39-3.31). The risk of composite neonatal morbidity with shoulder dystocia (12.2%) was also significantly higher than without shoulder dystocia (2.4%) (adjusted relative risk, 5.18; 95% confidence interval, 4.60-5.84). The most common neonatal morbidity was birth injury (adjusted relative risk, 5.39; 95% confidence interval, 4.71-6.17). The area under the curve for maternal characteristics to identify shoulder dystocia was 0.66 and it was 0.67 for intrapartum factors.

CONCLUSION

Although shoulder dystocia is unpredictable, the associated morbidity affects both mothers and newborns. The focus should be on concurrently averting the composite morbidity for the maternal-neonatal dyad with shoulder dystocia.

Predictors of neonatal brachial plexus palsy subsequent to resolution of shoulder dystocia

OBJECTIVE

The objective was to ascertain factors among deliveries complicated by shoulder dystocia (SD) and neonatal brachial plexus palsy (NBPP).

METHODS

At 11 hospitals, deliveries complicated by SD were identified. The inclusion criteria were vaginal delivery of non-anomalous, singleton at 34-42 weeks. Adjusted odds ratios (aOR) with 95% confidence intervals (CI) were calculated. Receiver operating characteristic (ROC) curves were created to evaluate the predictive value of the models for NBPP.

RESULTS

Of the 62,939 individuals who delivered vaginally, 1,134 (1.8%) had SD and met other inclusion criteria. Among the analytic cohort, 74 (6.5%) had NBPP. The factor known before delivery which was associated with NBPP was diabetes (aOR = 3.87; 95% CI = 2.13-7.01). After delivery, the three factors associated with NBPP were: (1) birthweight of at least 4000 g (aOR = 1.83; 95% CI = 1.05-3.20); (2) calling for help during the SD (aOR = 4.09, 95% CI = 2.29-7.30), and (3) the duration of SD ≥120 sec (aOR = 2.47, 95% CI = 1.30-4.69). The AUC under the ROC curve for these independent factors was 0.79 (95% CI = 0.77 - 0.82).

CONCLUSIONS

Few factors were identified that were associated with NBPP after SD, but they could not reliably predict which neonates will experience the complication.

Brain views that benefit from three-dimensional ultrasound

PURPOSE OF REVIEW

Fetal central nervous system malformations are among the most common congenital anomalies. Whereas simple axial views are sufficient for basic fetal brain examination, other important views are essential for a more detailed examination, which are sometimes challenging to obtain. Three-dimensional ultrasound can be helpful in obtaining standardized and reproducible images of many difficult fetal brain views. The aim of the present review is to explore the most recent evidence on the utility and technique of three-dimensional ultrasound in the examination of the fetal brain, with particular emphasis on the brain views that benefit from three-dimensional ultrasound.

RECENT FINDINGS

The article describes the various techniques of acquisition and analyses of three-dimensional ultrasound volumes of the fetal brain and their usefulness in the assessment of normal and abnormal fetal brain anatomy. Three-dimensional ultrasound has also permitted the application of many new technologies, such as artificial intelligence and deep machine learning. Recently, thanks to high-quality three-dimensional ultrasound, fetal cortical development can be assessed quantitatively and reliably.

SUMMARY

Three dimensional ultrasound can help as a complementary tool to two-dimensional ultrasound in the assessment of the fetal brain development and malformations. In addition, it paves the way for the application of promising technologies in the evaluation of fetal brain.

VIDEO ABSTRACT

A video summarizing the findings of the article. The video illustrates the various approaches and techniques applied for the examination of the fetal brain using three-dimensional ultrasound. Furthermore, the advantages and future perspectives of the application of three-dimensional ultrasound in the examination of the fetal brain are discussed, http://links.lww.com/COOG/A74.

Comparison of Multivariable Logistic Regression and Other Machine Learning Algorithms for Prognostic Prediction Studies in Pregnancy Care: Systematic Review and Meta-Analysis

BACKGROUND

Predictions in pregnancy care are complex because of interactions among multiple factors. Hence, pregnancy outcomes are not easily predicted by a single predictor using only one algorithm or modeling method.

OBJECTIVE

This study aims to review and compare the predictive performances between logistic regression (LR) and other machine learning algorithms for developing or validating a multivariable prognostic prediction model for pregnancy care to inform clinicians' decision making.

METHODS

Research articles from MEDLINE, Scopus, Web of Science, and Google Scholar were reviewed following several guidelines for a prognostic prediction study, including a risk of bias (ROB) assessment. We report the results based on the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. Studies were primarily framed as PICOTS (population, index, comparator, outcomes, timing, and setting): Population: men or women in procreative management, pregnant women, and fetuses or newborns; Index: multivariable prognostic prediction models using non-LR algorithms for risk classification to inform clinicians' decision making; Comparator: the models applying an LR; Outcomes: pregnancy-related outcomes of procreation or pregnancy outcomes for pregnant women and fetuses or newborns; Timing: pre-, inter-, and peripregnancy periods (predictors), at the pregnancy, delivery, and either puerperal or neonatal period (outcome), and either short- or long-term prognoses (time interval); and Setting: primary care or hospital. The results were synthesized by reporting study characteristics and ROBs and by random effects modeling of the difference of the logit area under the receiver operating characteristic curve of each non-LR model compared with the LR model for the same pregnancy outcomes. We also reported between-study heterogeneity by using τ2 and I2.

RESULTS

Of the 2093 records, we included 142 studies for the systematic review and 62 studies for a meta-analysis. Most prediction models used LR (92/142, 64.8%) and artificial neural networks (20/142, 14.1%) among non-LR algorithms. Only 16.9% (24/142) of studies had a low ROB. A total of 2 non-LR algorithms from low ROB studies significantly outperformed LR. The first algorithm was a random forest for preterm delivery (logit AUROC 2.51, 95% CI 1.49-3.53; I2=86%; τ2=0.77) and pre-eclampsia (logit AUROC 1.2, 95% CI 0.72-1.67; I2=75%; τ2=0.09). The second algorithm was gradient boosting for cesarean section (logit AUROC 2.26, 95% CI 1.39-3.13; I2=75%; τ2=0.43) and gestational diabetes (logit AUROC 1.03, 95% CI 0.69-1.37; I2=83%; τ2=0.07).

CONCLUSIONS

Prediction models with the best performances across studies were not necessarily those that used LR but also used random forest and gradient boosting that also performed well. We recommend a reanalysis of existing LR models for several pregnancy outcomes by comparing them with those algorithms that apply standard guidelines.

TRIAL REGISTRATION

PROSPERO (International Prospective Register of Systematic Reviews) CRD42019136106; https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=136106.

Implementation of machine learning models for the prediction of vaginal birth after cesarean delivery

OBJECTIVE

Accurate prediction of vaginal birth after cesarean is crucial for selecting women suitable for a trial of labor after cesarean (TOLAC). We sought to develop a machine learning (ML) model for prediction of TOLAC success and to compare its accuracy with that of the MFMU model.

METHODS

All consecutive singleton TOLAC deliveries from a tertiary academic medical center between February 2017 and December 2018 were included. We developed models using the following ML algorithms: random forest (RF), regularized regression (GLM), and eXtreme gradient-boosted decision trees (XGBoost). For developing the ML models, we disaggregated BMI into height and weight. Similarly, we disaggregated prior arrest of progression into prior arrest of dilatation and prior arrest of descent. We applied a nested cross-validation approach, using 100 random splits of the data to training (80%, 792 samples) and testing sets (20%, 197 samples). We used the area under the precision-recall curve (AUC-PR) as a measure of accuracy.

RESULTS

Nine hundred and eighty-nine TOLAC deliveries were included in the analysis with an observed TOLAC success rate of 85.6%. The AUC-PR in the RF, XGBoost and GLM models were 0.351±0.028, 0.350±0.028 and 0.336±0.024, respectively, compared to 0.325±0.067 for the MFMU-C. The algorithms performed significantly better than the MFMU-C (p-values = .0002, .0004, .0393 for RF, XGBoost, GLM respectively). In the XGBoost model, eight variables were sufficient for accurate prediction. In all ML models, previous vaginal delivery and height were among the three most important predictors of TOLAC success. Prior arrest of descent contributed to prediction more than prior arrest of dilatation, maternal height contributed more than weight.

CONCLUSION

All ML models performed significantly better than the MFMU-C. In the XGBoost model, eight variables were sufficient for accurate prediction. Prior arrest of descent and maternal height contribute to prediction more than prior arrest of dilation and maternal weight.

Introduction to artificial intelligence in ultrasound imaging in obstetrics and gynecology

Artificial intelligence (AI) uses data and algorithms to aim to draw conclusions that are as good as, or even better than, those drawn by humans. AI is already part of our daily life; it is behind face recognition technology, speech recognition in virtual assistants (such as Amazon Alexa, Apple's Siri, Google Assistant and Microsoft Cortana) and self-driving cars. AI software has been able to beat world champions in chess, Go and recently even Poker. Relevant to our community, it is a prominent source of innovation in healthcare, already helping to develop new drugs, support clinical decisions and provide quality assurance in radiology. The list of medical image-analysis AI applications with USA Food and Drug Administration or European Union (soon to fall under European Union Medical Device Regulation) approval is growing rapidly and covers diverse clinical needs, such as detection of arrhythmia using a smartwatch or automatic triage of critical imaging studies to the top of the radiologist's worklist. Deep learning, a leading tool of AI, performs particularly well in image pattern recognition and, therefore, can be of great benefit to doctors who rely heavily on images, such as sonologists, radiographers and pathologists. Although obstetric and gynecological ultrasound are two of the most commonly performed imaging studies, AI has had little impact on this field so far. Nevertheless, there is huge potential for AI to assist in repetitive ultrasound tasks, such as automatically identifying good-quality acquisitions and providing instant quality assurance. For this potential to thrive, interdisciplinary communication between AI developers and ultrasound professionals is necessary. In this article, we explore the fundamentals of medical imaging AI, from theory to applicability, and introduce some key terms to medical professionals in the field of ultrasound. We believe that wider knowledge of AI will help accelerate its integration into healthcare. © 2020 The Authors. Ultrasound in Obstetrics & Gynecology published by John Wiley & Sons Ltd on behalf of the International Society of Ultrasound in Obstetrics and Gynecology.

Does the Porter formula hold its promise? A weight estimation formula for macrosomic fetuses put to the test

Purpose

Estimating fetal weight using ultrasound measurements is an essential task in obstetrics departments. Most of the commonly used weight estimation formulas underestimate fetal weight when the actual birthweight exceeds 4000 g. Porter et al. published a specially designed formula in an attempt to improve detection rates for such macrosomic infants. In this study, we question the usefulness of the Porter formula in clinical practice and draw attention to some critical issues concerning the derivation of specialized formulas of this type.

Methods

A retrospective cohort study was carried out, including 4654 singleton pregnancies with a birthweight ≥ 3500 g, with ultrasound examinations performed within 14 days before delivery. Fetal weight estimations derived using the Porter and Hadlock formulas were compared.

Results

Of the macrosomic infants, 27.08% were identified by the Hadlock formula, with a false-positive rate of 4.60%. All macrosomic fetuses were detected using the Porter formula, with a false-positive rate of 100%; 99.96% of all weight estimations using the Porter formula fell within a range of 4300 g ± 10%. The Porter formula only provides macrosomic estimates.

Conclusions

The Porter formula does not succeed in distinguishing macrosomic from normal-weight fetuses. High-risk fetuses with a birthweight ≥ 4500 g in particular are not detected more precisely than with the Hadlock formula. For these reasons, we believe that the Porter formula should not be used in clinical practice. Newly derived weight estimation formulas for macrosomic fetuses must not be based solely on a macrosomic data set.

Electronic supplementary material

The online version of this article (10.1007/s00404-019-05410-7) contains supplementary material, which is available to authorized users.