Deep Learing for Sparse Domain Kalman Filtering with Applications on ECG Denoising and Motility Estimation.

OBJECTIVE: The reconstruction of an input based on a sparse combination of signals, known as sparse coding, has found widespread use in signal processing. In this work, the combination of sparse coding with Kalman filtering is explored and its potential is shown on two use-cases. METHODS: This work extends the Iterative Shrinkage and Thresholding Algorithm with a Kalman filter in the sparse domain. The resulting method may be implemented as a deep unfolded neural network and may be applied to any signal which has a sparse representation and a known or assumed relation between consecutive measurements. This method is evaluated on the use cases of noise reduction in the electrocardiogram (ECG) and the estimation of object motility. RESULTS: For ECG denoising, the proposed method achieved an improvement in Signal-to-Noise ratio of 18.6dB, which is comparable to state-of-the-art. In motility estimation, a correlation of 0.84 with ground truth simulations was found. CONCLUSION: The proposed method was shown to have advantages over sparse coding and Kalman filtering alone. Due to the low complexity and high generalizability of the proposed method, the implementation of context-specific knowledge or an extension to other applications can be readily made. SIGNIFICANCE: The presented Kalman-ISTA algorithm is a resource-efficient method combining the promise of both sparse coding and Kalman filtering, making it well-suited for various applications.

The impact of healthy pregnancy on features of heart rate variability and pulse wave morphology derived from wrist-worn photoplethysmography.

Due to the association between dysfunctional maternal autonomic regulation and pregnancy complications, tracking non-invasive features of autonomic regulation derived from wrist-worn photoplethysmography (PPG) measurements may allow for the early detection of deteriorations in maternal health. However, even though a plethora of these features-specifically, features describing heart rate variability (HRV) and the morphology of the PPG waveform (morphological features)-exist in the literature, it is unclear which of these may be valuable for tracking maternal health. As an initial step towards clarity, we compute comprehensive sets of HRV and morphological features from nighttime PPG measurements. From these, using logistic regression and stepwise forward feature elimination, we identify the features that best differentiate healthy pregnant women from non-pregnant women, since these likely capture physiological adaptations necessary for sustaining healthy pregnancy. Overall, morphological features were more valuable for discriminating between pregnant and non-pregnant women than HRV features (area under the receiver operating characteristics curve of 0.825 and 0.74, respectively), with the systolic pulse wave deterioration being the most valuable single feature, followed by mean heart rate (HR). Additionally, we stratified the analysis by sleep stages and found that using features calculated only from periods of deep sleep enhanced the differences between the two groups. In conclusion, we postulate that in addition to HRV features, morphological features may also be useful in tracking maternal health and suggest specific features to be included in future research concerning maternal health.

On the distinct differences in autonomic regulation between pregnant and non-pregnant women - a heart rate variability analysis.

Objective. Appropriate adaptation of the maternal autonomic nervous system to progressing gestation is essential to a healthy pregnancy. This is partly evidenced by the association between pregnancy complications and autonomic dysfunction. Therefore, assessing maternal heart rate variability (HRV)-a proxy measure for autonomic activity-may offer insights into maternal health, potentially enabling the early detection of complications. However, identifying abnormal maternal HRV requires a thorough understanding of normal maternal HRV. While HRV in women of childbearing age has been extensively investigated, less is known concerning HRV during pregnancy. Subsequently, we investigate the differences in HRV between healthy pregnant women and their non-pregnant counterparts.Approach. We use a comprehensive suite of HRV features (assessing sympathetic and parasympathetic activity, heart rate (HR) complexity, HR fragmentation, and autonomic responsiveness) to quantify HRV in large groups of healthy pregnant (n= 258) and non-pregnant women (n= 252). We compare the statistical significance and effect size of the potential differences between the groups.Main results. We find significantly increased sympathetic and decreased parasympathetic activity during healthy pregnancy, along with significantly attenuated autonomic responsiveness, which we hypothesize serves as a protective mechanism against sympathetic overactivity. HRV differences between these groups typically had a large effect size (Cohen'sd> 0.8), with the largest effect accompanying the significantly reduced HR complexity and altered sympathovagal balance observed in pregnancy (Cohen'sd> 1.2).Significance. Healthy pregnant women are autonomically distinct from their non-pregnant counterparts. Subsequently, assumptions based on HRV research in non-pregnant women cannot be readily translated to pregnant women.

Characterizing the effect of demographics, cardiorespiratory factors, and inter-subject variation on maternal heart rate variability in pregnancy with statistical modeling: a retrospective observational analysis.

Pregnancy complications are associated with insufficient adaptation of the maternal autonomic nervous system to the physiological demands of pregnancy. Consequently, assessing maternal heart rate variability (mHRV)-which reflects autonomic regulation-is a promising tool for detecting early deterioration in maternal health. However, before mHRV can be used to screen for complications, an understanding of the factors influencing mHRV during healthy pregnancy is needed. In this retrospective observational study, we develop regression models to unravel the effects of maternal demographics (age, body mass index (BMI), gestational age (GA), and parity), cardiorespiratory factors (heart rate and breathing rate), and inter-subject variation on mHRV. We develop these models using two datasets which are comprised of, respectively, single measurements in 290 healthy pregnant women and repeated measurements (median = 8) in 29 women with healthy pregnancies. Our most consequential finding is that between one-third and two-thirds of the variation in mHRV can be attributed to inter-subject variability. Additionally, median heart rate dominantly affects mHRV (p < 0.001), while BMI and parity have no effect. Moreover, we found that median breathing rate, age, and GA all impact mHRV (p < 0.05). These results suggest that personalized, long-term monitoring would be necessary for using mHRV for obstetric screening.

The electrical heart axis in fetuses with congenital heart disease, measured with non-invasive fetal electrocardiography.

OBJECTIVES: To determine if the electrical heart axis in different types of congenital heart defects (CHD) differs from that of a healthy cohort at mid-gestation. METHODS: Non-invasive fetal electrocardiography (NI-fECG) was performed in singleton pregnancies with suspected CHD between 16 and 30 weeks of gestation. The mean electrical heart axis (MEHA) was determined from the fetal vectorcardiogram after correction for fetal orientation. Descriptive statistics were used to determine the MEHA with corresponding 95% confidence intervals (CI) in the frontal plane of all fetuses with CHD and the following subgroups: conotruncal anomalies (CTA), atrioventricular septal defects (AVSD) and hypoplastic right heart syndrome (HRHS). The MEHA of the CHD fetuses as well as the subgroups was compared to the healthy control group using a spherically projected multivariate linear regression analysis. Discriminant analysis was applied to calculate the sensitivity and specificity of the electrical heart axis for CHD detection. RESULTS: The MEHA was determined in 127 fetuses. The MEHA was 83.0° (95% CI: 6.7°; 159.3°) in the total CHD group, and not significantly different from the control group (122.7° (95% CI: 101.7°; 143.6°). The MEHA was 105.6° (95% CI: 46.8°; 164.4°) in the CTA group (n = 54), -27.4° (95% CI: -118.6°; 63.9°) in the AVSD group (n = 9) and 26.0° (95% CI: -34.1°; 86.1°) in the HRHS group (n = 5). The MEHA of the AVSD and the HRHS subgroups were significantly different from the control group (resp. p = 0.04 and p = 0.02). The sensitivity and specificity of the MEHA for the diagnosis of CHD was 50.6% (95% CI 47.5% - 53.7%) and 60.1% (95% CI 57.1% - 63.1%) respectively. CONCLUSION: The MEHA alone does not discriminate between healthy fetuses and fetuses with CHD. However, the left-oriented electrical heart axis in fetuses with AVSD and HRHS was significantly different from the control group suggesting altered cardiac conduction along with the structural defect. TRIAL REGISTRATION: Clinical trial registration number: NL48535.015.14.

Maternal autonomic responsiveness is attenuated in healthy pregnancy: a phase rectified signal averaging analysis.

Autonomic regulation is essential in enabling a healthy pregnancy. In fact, several pregnancy complications are associated with autonomic dysfunction. Better understanding of the maternal autonomic state during healthy pregnancy may aid in the early detection of such complications. One aspect of autonomic regulation is autonomic responsiveness, which can by assessed by phase rectified signal averaging (PRSA). While other areas of research have found blunted physiological responses in pregnancy, this paper presents the first investigation of maternal autonomic responsiveness as assessed by PRSA. We find significantly reduced rates of responses, as well as an attenuated capacity for heart rate acceleration when comparing pregnant women to non-pregnant controls. We hypothesize that this attenuated autonomic control may serve to protect the mother against her imbalanced autonomic state, as increased sympathetic and decreased parasympathetic modulation accompany healthy pregnancies. Clinical Relevance- Maternal autonomic responsiveness is attenuated in pregnancy in comparison to non-pregnant women. Understanding maternal autonomic state not only improves our knowledge of gestational physiology but also forms the basis for the early detection of pregnancy complications associated with maternal autonomic dysfunction.

New possibilities for ST analysis - A post-hoc analysis on the Dutch STAN RCT.

BACKGROUND: The diagnostic value of ST analysis of the fetal electrocardiogram (fECG) during labor is uncertain. False alarms (ST events) may be explained by physiological variation of the fetal electrical heart axis. Adjusted ST events, based on a relative rather than an absolute rise from baseline, correct for this variation and may improve the diagnostic accuracy of ST analysis. AIMS: Determine the optimal cut-off for relative ST events in fECG to detect fetal metabolic acidosis. STUDY DESIGN: Post-hoc analysis on fECG tracings from the Dutch STAN trial (STAN+CTG branch). SUBJECTS: 1328 term singleton fetuses with scalp ECG tracing during labor, including 10 cases of metabolic acidosis. OUTCOME MEASURES: Cut-off value for relative ST events at the point closest to (0,1) in the receiver operating characteristic (ROC) curve with corresponding sensitivity and specificity. RESULTS: Relative baseline ST events had an optimal cut-off at an increment of 85% from baseline. Relative ST events had a sensitivity of 90% and specificity of 80%. CONCLUSIONS: Adjusting the current definition of ST events may improve ST analysis, making it independent of CTG interpretation.

A dilated inception CNN-LSTM network for fetal heart rate estimation.

Objective. Fetal heart rate (HR) monitoring is routinely used during pregnancy and labor to assess fetal well-being. The noninvasive fetal electrocardiogram (ECG), obtained by electrodes on the maternal abdomen, is a promising alternative to standard fetal monitoring. Subtraction of the maternal ECG from the abdominal measurements results in fetal ECG signals, in which the fetal HR can be determined typically through R-peak detection. However, the low signal-to-noise ratio and the nonstationary nature of the fetal ECG make R-peak detection a challenging task.Approach. We propose an alternative approach that instead of performing R-peak detection employs deep learning to directly determine the fetal HR from the extracted fetal ECG signals. We introduce a combination of dilated inception convolutional neural networks (CNN) with long short-term memory networks to capture both short-term and long-term temporal dynamics of the fetal HR. The robustness of the method is reinforced by a separate CNN-based classifier that estimates the reliability of the outcome.Main results. Our method achieved a positive percent agreement (within 10% of the actual fetal HR value) of 97.3% on a dataset recorded during labor and 99.6% on set-A of the 2013 Physionet/Computing in Cardiology Challenge exceeding top-performing state-of-the-art algorithms from the literature.Significance. The proposed method can potentially improve the accuracy and robustness of fetal HR extraction in clinical practice.

Deep Convolutional Long Short-Term Memory Network for Fetal Heart Rate Extraction.

Fetal electrocardiography is a valuable alternative to standard fetal monitoring. Suppression of the maternal electrocardiogram (ECG) in the abdominal measurements, results in fetal ECG signals, from which the fetal heart rate (HR) can be determined. This HR detection typically requires fetal R-peak detection, which is challenging, especially during low signal-to-noise ratio periods, caused for example by uterine activity. In this paper, we propose the combination of a convolutional neural network and a long short-term memory network that directly predicts the fetal HR from multichannel fetal ECG. The network is trained on a dataset, recorded during labor, while the performance of the method is evaluated both on a test dataset and on set-A of the 2013 Physionet /Computing in Cardiology Challenge. The algorithm achieved a positive percent agreement of 92.1% and 98.1% for the two datasets respectively, outperforming a top-performing state-of-the-art signal processing algorithm.

The fetal electrocardiogram to detect the effects of betamethasone on fetal heart rate variability.

BACKGROUND: Betamethasone is widely used to enhance fetal lung maturation in case of threatened preterm birth. Antenatal corticosteroids are known to reduce fetal heart rate variability (fHRV) in the days following administration. Since decreased fHRV is a marker for fetal distress, this transient decrease of fHRV can cause unnecessary medical intervention. AIM: To describe the effect of betamethasone on fHRV, by applying spectral analysis on non-invasive fetal electrocardiogram (fECG) recordings. STUDY DESIGN: Secondary analysis of a prospective cohort study. SUBJECTS: Women with a singleton pregnancy, at risk for preterm delivery and receiving betamethasone, admitted to the obstetric high care unit in the period from March 2013 until July 2016. OUTCOME MEASURES: The primary outcome measure was fHRV in both time- and frequency-domain. Secondary outcome measures included basal fetal heart rate (fHR) and fHR variance. FHRV parameters were then calculated separately for the quiet and active state. RESULTS: Following 68 inclusions, 22 patients remained with complete series of measurements and sufficient data quality. FHRV parameters and fHR showed a decrease on day 2 compared to day 1, significant for short-term variability and high-frequency power. Similar results were found when analyzing for separate behavioral states. The number of segments in quiet state increased during days 1 and 2. Normalized values showed no difference for all behavioral states. CONCLUSION: FHRV decreases on day 2 after betamethasone administration, while periods of fetal quiescence increase. No changes were found in the normalized values, indicating that the influence of autonomic modulation is minor. Clinical trial registration number NL43294.015.13.

Enhancement of low-quality fetal electrocardiogram based on time-sequenced adaptive filtering.

Extraction of a clean fetal electrocardiogram (ECG) from non-invasive abdominal recordings is one of the biggest challenges in fetal monitoring. An ECG allows for the interpretation of the electrical heart activity beyond the heart rate and heart rate variability. However, the low signal quality of the fetal ECG hinders the morphological analysis of its waveform in clinical practice. The time-sequenced adaptive filter has been proposed for performing optimal time-varying filtering of non-stationary signals having a recurring statistical character. In our study, the time-sequenced adaptive filter is applied to enhance the quality of multichannel fetal ECG after the maternal ECG is removed. To improve the performance of the filter in cases of low signal-to-noise ratio (SNR), we enhance the ECG reference signals by averaging consecutive ECG complexes. The performance of the proposed augmented time-sequenced adaptive filter is evaluated in both synthetic and real data from PhysioNet. This evaluation shows that the suggested algorithm clearly outperforms other ECG enhancement methods, in terms of uncovering the ECG waveform, even in cases with very low SNR. With the presented method, quality of the fetal ECG morphology can be enhanced to the extent that the ECG might be fit for use in clinical diagnostics. Graphical abstract The extracted fetal ECG signals from non-invasive abdominal recordings still contain a substantial amount of noise. The time-sequenced adaptive filter provides a relatively accurate estimate of the underlying fetal ECG signal when the quality of the reference channels is enhanced prior to filtering.

Detection rate of fetal distress using contraction-dependent fetal heart rate variability analysis.

OBJECTIVE: Monitoring of the fetal condition during labor is currently performed by cardiotocograpy (CTG). Despite the use of CTG in clinical practice, CTG interpretation suffers from a high inter- and intra-observer variability and a low specificity. In addition to CTG, analysis of fetal heart rate variability (HRV) has been shown to provide information on fetal distress. However, fetal HRV can be strongly influenced by uterine contractions, particularly during the second stage of labor. Therefore, the aim of this study is to examine if distinguishing contractions from rest periods can improve the detection rate of HRV features for fetal distress during the second stage of labor. APPROACH: We used a dataset of 100 recordings, containing 20 cases of fetuses with adverse outcome. The most informative HRV features were selected by a genetic algorithm and classification performance was evaluated using support vector machines. MAIN RESULTS: Classification performance of fetal heart rate segments closest to birth improved from a geometric mean of 70% to 79%. If the classifier was used to indicate fetal distress over time, the geometric mean at 15 minutes before birth improved from 60% to 72%. SIGNIFICANCE: Our results show that combining contraction-dependent HRV features with HRV features calculated over the entire fetal heart rate signal improves the detection rate of fetal distress.

A Fixed-Lag Kalman Smoother to Filter Power Line Interference in Electrocardiogram Recordings.

OBJECTIVE: Filtering power line interference (PLI) from electrocardiogram (ECG) recordings can lead to significant distortions of the ECG and mask clinically relevant features in ECG waveform morphology. The objective of this study is to filter PLI from ECG recordings with minimal distortion of the ECG waveform. METHODS: In this paper, we propose a fixed-lag Kalman smoother with adaptive noise estimation. The performance of this Kalman smoother in filtering PLI is compared to that of a fixed-bandwidth notch filter and several adaptive PLI filters that have been proposed in the literature. To evaluate the performance, we corrupted clean neonatal ECG recordings with various simulated PLI. Furthermore, examples are shown of filtering real PLI from an adult and a fetal ECG recording. RESULTS: The fixed-lag Kalman smoother outperforms other PLI filters in terms of step response settling time (improvements that range from 0.1 to 1 s) and signal-to-noise ratio (improvements that range from 17 to 23 dB). Our fixed-lag Kalman smoother can be used for semi real-time applications with a limited delay of 0.4 s. CONCLUSION AND SIGNIFICANCE: The fixed-lag Kalman smoother presented in this study outperforms other methods for filtering PLI and leads to minimal distortion of the ECG waveform.

Using uterine activity to improve fetal heart rate variability analysis for detection of asphyxia during labor.

During labor, uterine contractions can cause temporary oxygen deficiency for the fetus. In case of severe and prolonged oxygen deficiency this can lead to asphyxia. The currently used technique for detection of asphyxia, cardiotocography (CTG), suffers from a low specificity. Recent studies suggest that analysis of fetal heart rate variability (HRV) in addition to CTG can provide information on fetal distress. However, interpretation of fetal HRV during labor is difficult due to the influence of uterine contractions on fetal HRV. The aim of this study is therefore to investigate whether HRV features differ during contraction and rest periods, and whether these differences can improve the detection of asphyxia. To this end, a case-control study was performed, using 14 cases with asphyxia that were matched with 14 healthy fetuses. We did not find significant differences for individual HRV features when calculated over the fetal heart rate without separating contractions and rest periods (p > 0.30 for all HRV features). Separating contractions from rest periods did result in a significant difference. In particular the ratio between HRV features calculated during and outside contractions can improve discrimination between fetuses with and without asphyxia (p < 0.04 for three out of four ratio HRV features that were studied in this paper).

Selective heart rate variability analysis to account for uterine activity during labor and improve classification of fetal distress.

Cardiotocography (CTG) is currently the most often used technique for detection of fetal distress. Unfortunately, CTG has a poor specificity. Recent studies suggest that, in addition to CTG, information on fetal distress can be obtained from analysis of fetal heart rate variability (HRV). However, uterine contractions can strongly influence fetal HRV. The aim of this study is therefore to investigate whether HRV analysis for detection of fetal distress can be improved by distinguishing contractions from rest periods. Our results from feature selection indicate that HRV features calculated separately during contractions or during rest periods are more informative on fetal distress than HRV features that are calculated over the entire fetal heart rate. Furthermore, classification performance improved from a geometric mean of 69.0% to 79.6% when including the contraction-dependent HRV features, in addition to HRV features calculated over the entire fetal heart rate.

Motion artifacts in capacitive ECG measurements: reducing the combined effect of DC voltages and capacitance changes using an injection signal.

Capacitive electrodes are a promising alternative to the conventional adhesive electrodes for ECG measurements. They provide more comfort to the patient when integrated in everyday objects (e.g., beds or seats) for long-term monitoring. However, the application of capacitive sensors is limited by their high sensitivity to motion artifacts. For example, motion at the body-electrode interface causes variations of the coupling capacitance which, in the presence of a dc voltage across the coupling capacitor, create strong artifacts in the measurements. The origin, relevance, and reduction of this specific and important type of artifacts are studied here. An injection signal is exploited to track the variations of the coupling capacitance in real time. This information is then used by an identification scheme to estimate the artifacts and subtract them from the measurements. The method was evaluated in simulations, lab environments, and in a real-life recording on an adult's chest. For the type of artifact under study, a strong artifact reduction ranging from 40 dB for simulated data to 9 dB for a given real-life recording was achieved. The proposed method is automated, does not require any knowledge about the measurement system parameters, and provides an online estimate for the dc voltage across the coupling capacitor.

Unobtrusive ECG monitoring in the NICU using a capacitive sensing array.

The thin skin of preterm babies is easily damaged by adhesive electrodes, tapes, chest drains and needle-marks. The scars caused could be disfiguring or disabling to 10% of preterm newborns. Capacitive sensors present an attractive option for pervasively monitoring neonatal ECG, and can be embedded in a support system or even a garment worn by the neonate. This could improve comfort and reduce pain aiding better recovery as well as avoiding the scars caused by adhesive electrodes. In this work, we investigate the use of an array of capacitive sensors unobtrusively embedded in a mattress and used in a clinical environment for 15 preterm neonates. We also describe the analysis framework including the fusion of information from all sensors to provide a more accurate ECG signal. We propose a channel selection strategy as well as a method using physiological information to obtain a reliable ECG signal. When sensor coverage is well attained, results for both instantaneous heart rate and ECG signal shape analysis are very encouraging. The study also provides several insights on important factors affecting the results. These include the effect of textile type, number of layers, interferences (e.g. people walking by), motion severity and interventions. Incorporating this knowledge in the design of a capacitive sensing system would be crucial in ensuring that these sensors provide a reliable ECG signal when embedded in a neonatal support system.

ECG reconstruction based on the injection of a multi-frequency signal in capacitive measurement systems.

Many healthcare and lifestyle applications could benefit from capacitive measurement systems for unobtrusive ECG monitoring. However, a key technical challenge remains: the susceptibility of such systems to motion artifacts and common-mode interferences. With this in mind, we developed a novel method to reduce various types of artifacts present in capacitive ECG measurement systems. The objective is to perform ECG reconstruction and channel balancing in an automated and continuous manner. The proposed method consists of a) modeling the measurement system; b) specifically parameterizing the reconstruction equation; and c) adaptively estimating the parameters. A multi-frequency injection signal serves to estimate and track the variations of the different parameters of the reconstruction equation. A preliminary investigation on the validity of the method has been performed in both simulation and lab environment: the method shows benefits in terms of common-mode interference and motion artifact reduction, resulting in improved R-peak detection.

Reliability of spectral analysis of fetal heart rate variability.

Spectral analysis of fetal heart rate variability could provide information on fetal wellbeing. Unfortunately, fetal heart rate recordings are often contaminated by artifacts. Correction of these artifacts affects the outcome of spectral analysis, but it is currently unclear what level of artifact correction facilitates reliable spectral analysis. In this study, a method is presented that estimates the error in spectral powers due to artifact correction, based on the properties of the Continuous Wavelet Transformation. The results show that it is possible to estimate the error in spectral powers. The information about this error makes it possible for clinicians to assess the reliability of spectral analysis of fetal heart rate recordings that are contaminated by artifacts.

QRS classification and spatial combination for robust heart rate detection in low-quality fetal ECG recordings.

Non-invasive fetal electrocardiography (ECG) can be used for prolonged monitoring of the fetal heart rate (FHR). However, the signal-to-noise-ratio (SNR) of non-invasive ECG recordings is often insufficient for reliable detection of the FHR. To overcome this problem, source separation techniques can be used to enhance the fetal ECG. This study uses a physiology-based source separation (PBSS) technique that has already been demonstrated to outperform widely used blind source separation techniques. Despite the relatively good performance of PBSS in enhancing the fetal ECG, PBSS is still susceptible to artifacts. In this study an augmented PBSS technique is developed to reduce the influence of artifacts. The performance of the developed method is compared to PBSS on multi-channel non-invasive fetal ECG recordings. Based on this comparison, the developed method is shown to outperform PBSS for the enhancement of the fetal ECG.