New labor curves of dilation and station to improve the accuracy of predicting labor progress.

BACKGROUND: The diagnosis of failure to progress, the most common indication for intrapartum cesarean delivery, is based on the assessment of cervical dilation and station over time. Labor curves serve as references for expected changes in dilation and fetal descent. The labor curves of Friedman, Zhang et al, and others are based on time alone and derived from mothers with spontaneous labor onset. However, labor induction is now common, and clinicians also consider other factors when assessing labor progress. Labor curves that consider the use of labor induction and other factors that influence labor progress have the potential to be more accurate and closer to clinical decision-making. OBJECTIVE: This study aimed to compare the prediction errors of labor curves based on a single factor (time) or multiple clinically relevant factors using two modeling methods: mixed-effects regression, a standard statistical method, and Gaussian processes, a machine learning method. STUDY DESIGN: This was a longitudinal cohort study of changes in dilation and station based on data from 8022 births in nulliparous women with a live, singleton, vertex-presenting fetus ≥35 weeks of gestation with a vaginal delivery. New labor curves of dilation and station were generated with 10-fold cross-validation. External validation was performed using a geographically independent group. Model variables included time from the first examination in the 20 hours before delivery; dilation, effacement, and station recorded at the previous examination; cumulative contraction counts; and use of epidural anesthesia and labor induction. To assess model accuracy, differences between each model's predicted value and its corresponding observed value were calculated. These prediction errors were summarized using mean absolute error and root mean squared error statistics. RESULTS: Dilation curves based on multiple parameters were more accurate than those derived from time alone. The mean absolute error of the multifactor methods was better (lower) than those of the single-factor methods (0.826 cm [95% confidence interval, 0.820-0.832] for the multifactor machine learning and 0.893 cm [95% confidence interval, 0.885-0.901] for the multifactor mixed-effects method and 2.122 cm [95% confidence interval, 2.108-2.136] for the single-factor methods; P<.0001 for both comparisons). The root mean squared errors of the multifactor methods were also better (lower) than those of the single-factor methods (1.126 cm [95% confidence interval, 1.118-1.133] for the machine learning [P<.0001] and 1.172 cm [95% confidence interval, 1.164-1.181] for the mixed-effects methods and 2.504 cm [95% confidence interval, 2.487-2.521] for the single-factor [P<.0001 for both comparisons]). The multifactor machine learning dilation models showed small but statistically significant improvements in accuracy compared to the mixed-effects regression models (P<.0001). The multifactor machine learning method produced a curve of descent with a mean absolute error of 0.512 cm (95% confidence interval, 0.509-0.515) and a root mean squared error of 0.660 cm (95% confidence interval, 0.655-0.666). External validation using independent data produced similar findings. CONCLUSION: Cervical dilation models based on multiple clinically relevant parameters showed improved (lower) prediction errors compared to models based on time alone. The mean prediction errors were reduced by more than 50%. A more accurate assessment of departure from expected dilation and station may help clinicians optimize intrapartum management.

The evolution of the labor curve and its implications for clinical practice: the relationship between cervical dilation, station, and time during labor.

The assessment of labor progress is germane to every woman in labor. Two labor disorders-arrest of dilation and arrest of descent-are the primary indications for surgery in close to 50% of all intrapartum cesarean deliveries and are often contributing indications for cesarean deliveries for fetal heart rate abnormalities. Beginning in 1954, the assessment of labor progress was transformed by Friedman. He published a series of seminal works describing the relationship between cervical dilation, station of the presenting part, and time. He proposed nomenclature for the classification of labor disorders. Generations of obstetricians used this terminology and normal labor curves to determine expected rates of dilation and fetal descent and to decide when intervention was required. The analysis of labor progress presents many mathematical challenges. Clinical measurements of dilation and station are imprecise and prone to variation, especially for inexperienced observers. Many interrelated factors influence how the cervix dilates and how the fetus descends. There is substantial variability in when data collection begins and in the frequency of examinations. Statistical methods to account for these issues have advanced considerably in recent decades. In parallel, there is growing recognition among clinicians of the limitations of using time alone to assess progress in cervical dilation in labor. There is wide variation in the patterns of dilation over time and most labors do not follow an average dilation curve. Reliable assessment of labor progression is important because uncertainty leads to both over-use and under-use of cesarean delivery and neither of these extremes are desirable. This review traces the evolution of labor curves, describes how limitations are being addressed to reduce uncertainty and to improve the assessment of labor progression using modern statistical techniques and multi-dimensional data, and discusses the implications for obstetrical practice.

The limits of electronic fetal heart rate monitoring in the prevention of neonatal metabolic acidemia.

BACKGROUND: Despite intensive efforts directed at initial training in fetal heart rate interpretation, continuing medical education, board certification/recertification, team training, and the development of specific protocols for the management of abnormal fetal heart rate patterns, the goals of consistently preventing hypoxia-induced fetal metabolic acidemia and neurologic injury remain elusive. OBJECTIVE: The purpose of this study was to validate a recently published algorithm for the management of category II fetal heart rate tracings, to examine reasons for the birth of infants with significant metabolic acidemia despite the use of electronic fetal heart rate monitoring, and to examine critically the limits of electronic fetal heart rate monitoring in the prevention of neonatal metabolic acidemia. STUDY DESIGN: The potential performance of electronic fetal heart rate monitoring under ideal circumstances was evaluated in an outcomes-blinded examination fetal heart rate tracing of infants with metabolic acidemia at birth (base deficit, >12) and matched control infants (base deficit, <8) under the following conditions: (1) expert primary interpretation, (2) use of a published algorithm that was developed and endorsed by a large group of national experts, (3) assumption of a 30-minute period of evaluation for noncritical category II fetal heart rate tracings, followed by delivery within 30 minutes, (4) evaluation without the need to provide patient care simultaneously, and (5) comparison of results under these circumstances with those achieved in actual clinical practice. RESULTS: During the study period, 120 infants were identified with an arterial cord blood base deficit of >12 mM/L. Matched control infants were not demographically different from subjects. In actual practice, operative intervention on the basis of an abnormal fetal heart rate tracings occurred in 36 of 120 fetuses (30.0%) with metabolic acidemia. Based on expert, algorithm-assisted reviews, 55 of 120 patients with acidemia (45.8%) were judged to need operative intervention for abnormal fetal heart rate tracings. This difference was significant (P=.016). In infants who were born with a base deficit of >12 mM/L in which blinded, algorithm-assisted expert review indicated the need for operative delivery, the decision for delivery would have been made an average of 131 minutes before the actual delivery. The rate of expert intervention for fetal heart rate concerns in the nonacidemic control group (22/120; 18.3%) was similar to the actual intervention rate (23/120; 19.2%; P=1.0) Expert review did not mandate earlier delivery in 65 of 120 patients with metabolic acidemia. The primary features of these 65 cases included the occurrence of sentinel events with prolonged deceleration just before delivery, the rapid deterioration of nonemergent category II fetal heart rate tracings before realistic time frames for recognition and intervention, and the failure of recognized fetal heart rate patterns such as variability to identify metabolic acidemia. CONCLUSIONS: Expert, algorithm-assisted fetal heart rate interpretation has the potential to improve standard clinical performance by facilitating significantly earlier recognition of some tracings that are associated with metabolic acidemia without increasing the rate of operative intervention. However, this improvement is modest. Of infants who are born with metabolic acidemia, only approximately one-half potentially could be identified and have delivery expedited even under ideal circumstances, which are probably not realistic in current US practice. This represents the limits of electronic fetal heart rate monitoring performance. Additional technologies will be necessary if the goal of the prevention of neonatal metabolic acidemia is to be realized.

Assessing first-stage labor progression and its relationship to complications.

BACKGROUND: New labor curves have challenged the traditional understanding of the general pattern of dilation and descent in labor. They also revealed wide variation in the time to advance in dilation. An interval of arrest such as 4 hours did not fall beyond normal limits until dilation had reached 6 cm. Thus, the American College of Obstetricians and Gynecologists/Society for Maternal-Fetal Medicine first-stage arrest criteria, based in part on these findings, are applicable only in late labor. The wide range of time to dilate is unavoidable because cervical dilation has neither a precise nor direct relationship to time. Newer statistical techniques (multifactorial models) can improve precision by incorporating several factors that are related directly to labor progress. At each examination, the calculations adapt to the mother's current labor conditions. They produce a quantitative assessment that is expressed in percentiles. Low percentiles indicate potentially problematic labor progression. OBJECTIVE: The purpose of this study was to assess the relationship between first-stage labor progress- and labor-related complications with the use of 2 different assessment methods. The first method was based on arrest of dilation definitions. The other method used percentile rankings of dilation or station based on adaptive multifactorial models. STUDY DESIGN: We included all 4703 cephalic-presenting, term, singleton births with electronic fetal monitoring and cord gases at 2 academic community referral hospitals in 2012 and 2013. We assessed electronic data for route of delivery, all dilation and station examinations, newborn infant status, electronic fetal monitoring tracings, and cord blood gases. The labor-related complication groups included 272 women with cesarean delivery for first-stage arrest, 558 with cesarean delivery for fetal heart rate concerns, 178 with obstetric hemorrhage, and 237 with neonatal depression, which left 3004 women in the spontaneous vaginal birth group. Receiver operating characteristic curves were constructed for each assessment method by measurement of the sensitivity for each complication vs the false-positive rate in the normal reference group. RESULTS: The duration of arrest at ≥6 cm dilation showed poor levels of discrimination for the cesarean delivery interventions (area under the curve, 0.55-0.65; P < .01) and no significant relationship to hemorrhage or neonatal depression. The dilation and station percentiles showed high discrimination for the cesarean delivery-related outcomes (area under the curve, 0.78-0.93; P < .01) and low discrimination for the clinical outcomes of hemorrhage and neonatal depression (area under the curve, 0.58-0.61; P < .01). CONCLUSIONS: Duration of arrest of dilation at ≥6 cm showed little or no discrimination for any of the complications. In comparison, percentile rankings that were based on the adaptive multifactorial models showed much higher discrimination for cesarean delivery interventions and better, but low discrimination for hemorrhage. Adaptive multifactorial models present a different method to assess labor progress. Rather than "pass/fail" criteria that are applicable only to dilation in late labor, they produce percentile rankings, assess 2 essential processes for vaginal birth (dilation and descent), and can be applied from 3 cm onward. Given the limitations of labor-progress assessment based solely on the passage of time and because of the extreme variation in decision-making for cesarean delivery for labor disorders, the types of mathematic analyses that are described in this article are logical and promising steps to help standardize labor assessment.

Data-Driven Insights into Labor Progression with Gaussian Processes.

Clinicians routinely perform pelvic examinations to assess the progress of labor. Clinical guidelines to interpret these examinations, using time-based models of cervical dilation, are not always followed and have not contributed to reducing cesarean-section rates. We present a novel Gaussian process model of labor progress, suitable for real-time use, that predicts cervical dilation and fetal station based on clinically relevant predictors available from the pelvic exam and cardiotocography. We show that the model is more accurate than a statistical approach using a mixed-effects model. In addition, it provides confidence estimates on the prediction, calibrated to the specific delivery. Finally, we show that predicting both dilation and station with a single Gaussian process model is more accurate than two separate models with single predictions.

Construction of a comprehensive fetal monitoring database for the study of perinatal hypoxic ischemic encephalopathy.

This article describes the methods used to build a large-scale database of more than 250,000 electronic fetal monitoring (EFM) records linked to a comprehensive set of clinical information about the infant, the mother, the pregnancy, labor, and outcome. The database can be used to investigate how birth outcome is related to clinical and EFM features. The main steps involved in building the database were: (1) Acquiring the raw EFM recording and clinical records for each birth. (2) Assigning each birth to an objectively defined outcome class that included normal, acidosis, and hypoxic-ischemic encephalopathy. (3) Removing all personal health information from the EFM recordings and clinical records. (4) Preprocessing the deidentified EFM records to eliminate duplicates, reformat the signals, combine signals from different sensors, and bridge gaps to generate signals in a format that can be readily analyzed. (5) Post-processing the repaired EFM recordings to extract key features of the fetal heart rate, uterine activity, and their relations. (6) Populating a database that links the clinical information, EFM records, and EFM features to support easy querying and retrieval. •A multi-step process is required to build a comprehensive database linking electronic temporal fetal monitoring signals to a comprehensive set of clinical information about the infant, the mother, the pregnancy, labor, and outcome.•The current database documents more than 250,000 births including almost 4,000 acidosis and 400 HIE cases. This represents more than 80% of the births that occurred in 15 Northern California Kaiser Permanente Hospitals between 2011-2019. This is a valuable resource for studying the factors predictive of outcome.•The signal processing code and schemas for the database are freely available. The database will not be permitted to leave Kaiser firewalls, but a process is in place to allow interested investigators to access it.

New perspectives in electronic fetal surveillance.

Despite its recognized limitations, fetal heart rate monitoring is a mainstay of intrapartum care. Although the basic technology in standard electronic fetal monitors has changed little in recent decades, clinical behavior in response to heart rate monitoring has changed considerably. In addition to clearly defined nomenclature and clinical guidelines, there is an increased awareness that environmental and human factors can impair clinical judgment, resulting in delayed intervention and, consequently, birth-related injury. This review examines three essential steps that affect clinical outcome: (1) signal acquisition, (2) associations with physiological outcome, and (3) clinical intervention. Only the third step is directly responsible for changing clinical outcome. However, timely initiation of interventions is dependent upon the second step, which is dependent upon the fi rst step. Thus, deficiencies at each step tend to accumulate and contribute to the worsening of overall clinical outcome. This review article summarizes advances occurring at each step. The synergy and convergence of innovations in engineering, mathematics, and behavioral science shows considerable promise in intrapartum fetal surveillance.

Timely Detection of Infants at Risk of Intrapartum Acidosis and Hypoxic-Ischemic Encephalopathy Using Cardiotocography.

This work aims to improve the intrapartum detection of fetuses with an increased risk of developing fetal acidosis or hypoxic-ischemic encephalopathy (HIE) using fetal heart rate (FHR) and uterine pressure (UP) signals. Our study population comprised 40,831 term births divided into 3 classes based on umbilical cord or early neonatal blood gas assessments: 374 with verified HIE, 3,047 with acidosis but no encephalopathy and 37,410 healthy babies with normal gases. We developed an intervention recommendation system based on a random forest classifier. The classifier was trained using classical and novel features extracted electronically from 20-minute epochs of FHR and UP. Then, using the predictions of the classifier on each epoch, we designed a decision rule to determine when to recommended intervention. Compared to the Caesarean rates in each study group, our system identified an additional 5.68% of babies who developed HIE (54.55% vs 60.23%, p < 0.01) with a specific alert threshold. Importantly, about 75% of these recommendations were made more than 200 minutes before birth. In the acidosis group, the system identified an additional 17.44% (37.15% vs 54.59%, p < 0.01) and about 2/3 of these recommendations were made more than 200 minutes before birth. Compared to the Caesarean rate in the healthy group, the associated false positive rate was increased by 1.07% (38.80% vs 39.87%, p<0.01).Clinical Relevance- This method recommended intervention in more babies affected by acidosis or HIE, than the intervention rate observed in practice and most often did so 200 minutes before delivery. This was early enough to expect that interventions would have clinical benefit and reduce the rate of HIE. Given the high burden associated with HIE, this would justify the marginal increase in the normal Cesarean rate.

Accounting for Nulliparity in the Prediction of Hypoxic-Ischemic Encephalopathy Using Cardiotocography.

Nulliparous pregnancies, those where the mother has not previously given birth, are associated with longer labors and hence expose the fetus to more contractions and other adverse intrapartum conditions such as chorioamnionitis. The objective of the present study was to test if accounting for nulliparity could improve the detection of fetuses at increased risk of developing hypoxic-ischemic encephalopathy (HIE). During labor, clinicians assess the fetal heart rate and uterine pressure signals to identify fetuses at risk of developing HIE. In this study, we performed random forest classification using fetal heart rate and uterine pressure features from 40,831 births, including 374 that developed HIE. We analyzed a two-path classification approach that analyzed separately the fetuses from nulliparous and multiparous mothers, and a one-path classification approach that included the clinical variable for nulliparity as a classification feature. We compared these two approaches to a one-path classifier that had no information about the parity of the mothers. We also compared our results to the rate of Caesarean deliveries in each group, which is used clinically to interrupt the progression towards HIE. All the classifiers detected more fetuses that developed HIE than the observed Caesarean rate, but accounting for nulliparity did not improve performance.

Arrhythmia classification of 12-lead and reduced-lead electrocardiograms via recurrent networks, scattering, and phase harmonic correlation.

We describe an automatic classifier of arrhythmias based on 12-lead and reduced-lead electrocardiograms. Our classifier comprises four modules: scattering transform (ST), phase harmonic correlation (PHC), depthwise separable convolutions (DSC), and a long short-term memory (LSTM) network. It is trained on PhysioNet/Computing in Cardiology Challenge 2021 data. The ST captures short-term temporal ECG modulations while the PHC characterizes the phase dependence of coherent ECG components. Both reduce the sampling rate to a few samples per typical heart beat. We pass the output of the ST and PHC to a depthwise-separable convolution layer (DSC) which combines lead responses separately for each ST or PHC coefficient and then combines resulting values across all coefficients. At a deeper level, two LSTM layers integrate local variations of the input over long time scales. We train in an end-to-end fashion as a multilabel classification problem with a normal and 25 arrhythmia classes. Lastly, we use canonical correlation analysis (CCA) for transfer learning from 12-lead ST and PHC representations to reduced-lead ones. After local cross-validation on the public data from the challenge, our team 'BitScattered' achieved the following results: 0.682 ± 0.0095 for 12-lead; 0.666 ± 0.0257 for 6-lead; 0.674 ± 0.0185 for 4-lead; 0.661 ± 0.0098 for 3-lead; and 0.662 ± 0.0151 for 2-lead.

Multi-Chain Semi-Markov Analysis of Intrapartum Cardiotocography.

Visual assessment of the evolution of fetal heart rate (FHR) and uterine pressure (UP) patterns is the standard of care in the intrapartum period. Unfortunately, this assessment has high levels of intra- and inter-observer variability. This study processed and analyzed FHR and UP patterns using computerized pattern recognition tools. The goal was to evaluate differences in FHR and UP patterns between fetuses with normal outcomes and those who developed hypoxic-ischemic encephalopathy (HIE). For this purpose, we modeled the sequence of FHR patterns and uterine contractions using Multi-Chain Semi-Markov models (MCSMMs). These models estimate the probability of transitioning between FHR or UP patterns and the dwell time of each pattern. Our results showed that in comparison to the control group, the HIE group had: (1) more frequent uterine contractions during the last 12 hours before birth; (2) more frequent FHR decelerations during the last 12 hours before birth; (3) longer decelerations during the last eight hours before birth; and (4) shorter baseline durations during the last five hours before birth. These results demonstrate that the fetuses in the HIE group were subject to a more stressful environment than those in the normal group. Clinical Relevance- Our results revealed statistically significant differences in FHR/UP patterns between the normal and HIE groups in the hours before birth. This indicates that features derived using MCSMMs may be useful in a machine learning framework to detect infants at increased risk of developing HIE allowing preventive interventions.

Graded classification of fetal heart rate tracings: association with neonatal metabolic acidosis and neurologic morbidity.

OBJECTIVE: The objective of the study was to measure the performance of a 5-tier, color-coded graded classification of electronic fetal monitoring (EFM). STUDY DESIGN: We used specialized software to analyze and categorize 7416 hours of EFM from term pregnancies. We measured how often and for how long each of the color-coded levels appeared in 3 groups of babies: (A) 60 babies with neonatal encephalopathy (NE) and umbilical artery base deficit (BD) levels were greater than 12 mmol/L; (I) 280 babies without NE but with BD greater than 12 mmol/L; and (N) 2132 babies with normal gases. RESULTS: The frequency and duration of EFM abnormalities considered more severe in the classification method were highest in group A and lowest in group N. Detecting an equivalent percentage of cases with adverse outcomes required only minutes spent with marked EFM abnormalities compared with much longer periods with lesser abnormalities. CONCLUSION: Both degree and duration of tracing abnormality are related to outcome. We present empirical data quantifying that relationship in a systematic fashion.

Information theoretic measures of perinatal cardiotocography synchronization.

We examined the use of bivariate mutual information (MI) and its conditional variant transfer entropy (TE) to address synchronization of perinatal uterine pressure (UP) and fetal heart rate (FHR). We used a nearest-neighbour based Kraskov entropy estimator, suitable to the non-Gaussian distributions of the UP and FHR signals. Moreover, the estimates were robust to noise by use of surrogate data testing. Estimating degree of synchronicity and UP-FHR delay length is useful since they are physiological correlates to fetal hypoxia. Mutual information of the UP-FHR discriminated normal and pathological fetuses early (160 min before delivery) and discriminated normal and metabolic acidotic fetuses slightly later (110 min before delivery), with higher mutual information for progressively pathological classes. The delay in mutual information transfer was also discriminating in the last 50 min of labour. Transfer entropy discriminated normal and pathological cases 110 min before delivery with lower TE values and longer information transfer delays in pathological cases, to our knowledge, the first report of this phenomena in the literature.

Estimating risk of severe neonatal morbidity in preterm births under 32 weeks of gestation.

Background: A large recent study analyzed the relationship between multiple factors and neonatal outcome and in preterm births. Study variables included the reason for admission, indication for delivery, optimal steroid use, gestational age, and other potential prognostic factors. Using stepwise multivariable analysis, the only two variables independently associated with serious neonatal morbidity were gestational age and the presence of suspected intrauterine growth restriction as a reason for admission. This finding was surprising given the beneficial effects of antenatal steroids and hazards associated with some causes of preterm birth. Multivariable logistic regression techniques have limitations. Without testing for multiple interactions, linear regression will identify only individual factors with the strongest independent relationship to the outcome for the entire study group. There may not be a single "best set" of risk factors or one set that applies equally well to all subgroups. In contrast, machine learning techniques find the most predictive groupings of factors based on their frequency and strength of association, with no attempt to identify independence and no assumptions about linear relationships.Objective: To determine if machine learning techniques would identify specific clusters of conditions with different probability estimates for severe neonatal morbidity and to compare these findings to those based on the original multivariable analysis.Materials and methods: This was a secondary analysis of data collected in a multicenter, prospective study on all admissions to the neonatal intensive care unit between 2013 and 2015 in 10 hospitals. We included all patients with a singleton, stillborn, or live newborns, with a gestational age between 23 0/7 and 31 6/7 week. The composite endpoint, severe neonatal morbidity, defined by the presence of any of five outcomes: death, grade 3 or 4 intraventricular hemorrhage (IVH), and ≥28 days on ventilator, periventricular leukomalacia (PVL), or stage III necrotizing enterocolitis (NEC), was present in 238 of the 1039 study patients. We studied five explanatory variables: maternal age, parity, gestational age, admission reason, and status with respect to antenatal steroid administration. We concentrated on Classification and Regression Trees because the resulting structure defines clusters of risk factors that often bear resemblance to clinical reasoning. Model performance was measured using area under the receiver-operator characteristic curves (AUC) based on 10 repetitions of 10-fold cross-validation.Results: A hybrid technique using a combination of logistic regression and Classification and Regression Trees had a mean cross-validated AUC of 0.853. A selected point on its receiver-operator characteristic (ROC) curve corresponding to a sensitivity of 81% was associated with a specificity of 76%. Rather than a single curve representing the general relationship between gestational age and severe morbidity, this technique found seven clusters with distinct curves. Abnormal fetal testing as a reason for admission with or without growth restriction and incomplete steroid administration would place a 20-year-old patient on the highest risk curve.Conclusions: Using a relatively small database and a few simple factors known before birth it is possible to produce a more tailored estimate of the risk for severe neonatal morbidity on which clinicians can superimpose their medical judgment, experience, and intuition.

Hybrid scattering-LSTM networks for automated detection of sleep arousals.

OBJECTIVE: Early detection of sleep arousal in polysomnographic (PSG) signals is crucial for monitoring or diagnosing sleep disorders and reducing the risk of further complications, including heart disease and blood pressure fluctuations. APPROACH: In this paper, we present a new automatic detector of non-apnea arousal regions in multichannel PSG recordings. This detector cascades four different modules: a second-order scattering transform (ST) with Morlet wavelets; depthwise-separable convolutional layers; bidirectional long short-term memory (BiLSTM) layers; and dense layers. While the first two are shared across all channels, the latter two operate in a multichannel formulation. Following a deep learning paradigm, the whole architecture is trained in an end-to-end fashion in order to optimize two objectives: the detection of arousal onset and offset, and the classification of the type of arousal. Main results and Significance: The novelty of the approach is three-fold: it is the first use of a hybrid ST-BiLSTM network with biomedical signals; it captures frequency information lower (0.1 Hz) than the detection sampling rate (0.5 Hz); and it requires no explicit mechanism to overcome class imbalance in the data. In the follow-up phase of the 2018 PhysioNet/CinC Challenge the proposed architecture achieved a state-of-the-art area under the precision-recall curve (AUPRC) of 0.50 on the hidden test data, tied for the second-highest official result overall.

Ensembling convolutional and long short-term memory networks for electrocardiogram arrhythmia detection.

OBJECTIVE: Atrial fibrillation is a common type of heart rhythm abnormality caused by a problem with the heart's electrical system. Early detection of this disease has important implications for stroke prevention and management. Our objective is to construct an intelligent tool that assists cardiologists in identifying automatically cardiac arrhythmias and noise in electrocardiogram (ECG) recordings. APPROACH: Our base deep classifier combined a convolutional neural network (CNNs) and a sequence of long short-term memory units, with pooling, dropout and normalization techniques to improve their accuracy. The network predicted a classification at every 18th input sample and the final prediction was selected for classification. Ten standalone models that used our base classifier architecture were first cross-validated separately on 90% of the PhysioNet/CinC Challenge 2017 dataset and then tested on 10%. An ensemble classifier selected the label of the best average probability from the ten sub-models to improve prediction quality. MAIN RESULTS: Our original result submitted to the challenge gave a mean F1-measure of 80%. The new proposed method improved the test score to 82%, which was tied for the third-highest score in the follow-up phase of the challenge. SIGNIFICANCE: Without employing a time-consuming feature engineering step, the ensemble classifier trained with this architecture provided a robust solution to the problem of detecting cardiac arrhythmia from noisy ECG signals. In addition, interpretation of the classifier by inspection of its network parameters and predictions revealed what aspects of the ECG signal the classifier considered most discriminating.

Clinical associations with uterine tachysystole.

OBJECTIVES: To determine the incidence of uterine tachysystole (UT) and its association with neonatal depression or metabolic acidemia (DEP). METHODS: This retrospective study comprised all 6234 women at ≥ 37 weeks' gestation who were monitored during the last 4 hours of tracings before birth in an academic community hospital. DEP was defined by an umbilical artery base deficit value ≥ 10 mmol/L or a 5-minute Apgar ≤ 6 and included 77 births. UT was defined by >15 contractions in 30 minutes. RESULTS: The overall incidence of UT was 18.3% (1139/6234). In 4.2% (260/6234) UT persisted for >60 min. The rate of UT was similar in births with DEP (14.3%, 11/77) compared to those without DEP (18.3%, 1128/6157; p=0.45). In births with UT, only 1.0% (11/1139) developed DEP. The DEP group had more decelerations at almost every level of contractions and a higher cesarean rate of 49.4% (38/77) compared to 24.0% (1468/6124); p=<0.001 in the group without DEP. CONCLUSIONS: UT was common, occasionally prolonged and almost always benign. Fetuses with DEP had no more UT than those without DEP. Many babies with DEP declared their vulnerability with decelerations at contraction rates below UT levels and the great majority of them never experienced UT.

Subspace detection of the impulse response function from intra-partum cardiotocography.

Recording of maternal uterine pressure (UP) and fetal heart rate (FHR) during labour and delivery is a procedure referred to as cardiotocography (CTG). We model this as an input-output system to estimate its dynamics in terms of an impulse response function (IRF). CTG data is very noisy and missing data are common. In this paper, we identify the models using subspace methods, which incorporate noise-suppression and permit the use of non-contiguous data. Using contiguous data, the subspace method performed better than linear regression; more of the 57 CTG pathological records in our database were modelled (30 vs. 26). Allowing non-contiguous data, even more pathological records were modelled with this approach (49). Furthermore, the models were discriminating; compared to linear regression, the IRF gain showed statistically significant differences more often between normal and pathological records (in 15/18 vs. 10/18 epochs) over the final three hours of labour.

Classification of normal and hypoxic fetuses from systems modeling of intrapartum cardiotocography.

Recording of maternal uterine pressure (UP) and fetal heart rate (FHR) during labor and delivery is a procedure referred to as cardiotocography. We modeled this signal pair as an input-output system using a system identification approach to estimate their dynamic relation in terms of an impulse response function. We also modeled FHR baseline with a linear fit and FHR variability unrelated to UP using the power spectral density, computed from an auto-regressive model. Using a perinatal database of normal and pathological cases, we trained support-vector-machine classifiers with feature sets from these models. We used the classification in a detection process. We obtained the best results with a detector that combined the decisions of classifiers using both feature sets. It detected half of the pathological cases, with very few false positives (7.5%), 1 h and 40 min before delivery. This would leave sufficient time for an appropriate clinical response. These results clearly demonstrate the utility of our method for the early detection of cases needing clinical intervention.

Identification of the dynamic relationship between intrapartum uterine pressure and fetal heart rate for normal and hypoxic fetuses.

Labor and delivery are routinely monitored electronically with sensors that measure and record maternal uterine pressure (UP) and fetal heart rate (FHR), a procedure referred to as cardiotocography (CTG). Delay or failure to recognize abnormal patterns in these recordings can result in a failure to prevent fetal injury. We address the challenging problem of interpreting intrapartum CTG in a novel way by modeling the dynamic relationship between UP (as an input) and FHR (as an output). We use a nonparametric approach to estimate the dynamics in terms of an impulse response function (IRF). We apply singular value decomposition to suppress noise, IRF delay, and memory estimation to identify the temporal extent of the response and surrogate testing to assess model significance. We construct models for a database of CTG recordings labeled by outcome, and compare the models during the last 3 h of labor as well as across outcome classes. The results demonstrate that the UP-FHR dynamics can be successfully modeled as an input-output system. Models for pathological cases had stronger, more delayed, and more predictable responses than those for normal cases. In addition, the models evolved in time, reflecting a clinically plausible evolution of the fetal state due to the stress of labor.