Sage Journals: Discover world-class research

Abstract

With the adoption of the two-child policy, there has been a large increase in women of older maternal and high-risk pregnant women. So, it is necessary to analyze the health status of women in the late pregnancy on time. To analyze the effect on using remote fetal monitoring on women in the late pregnancy, we selected women in the late stage of pregnancy in our hospital as research subjects. They were randomly divided into two groups: the experimental group, which engaged in remote fetal monitoring, and the control group, which adopted traditional cardiac monitoring. In order to get more effective data, we used the Kalman filter and audio repair algorithms to preprocess the collected data. During follow-up observation, we compared the two groups using neonatal cardiac monitoring by employing the non-stress test and observed the occurrence of neonatal asphyxia. The incidence of neonatal abnormal non-stress test in the experimental group and the control group was 33.6% and 17.3%, respectively; the difference was statistically significant (p < 0.05). The incidence of neonatal asphyxia in the experimental group was 12.5%, which was significantly lower than in the control group (30%; p < 0.05). We have found that women in the late stage of pregnancy who adopted remote fetal monitoring could detect abnormal non-stress test earlier and thus increase in the detection of rate of neonatal asphyxia.

Keywords

Late pregnancy remote fetal monitoring Kalman filter repair algorithm

Introduction

With the adoption of the two-child policy,¹ there has been a large increase in women of advanced reproductive age and high-risk pregnant women, for whom concurrent medical problems are increasingly serious. This causes adverse outcomes during the perinatal period and affects the health of the postpartum mother and baby. Previous studies have shown that pregnancy complications, birth defects, and other health problems are the result of the combination of long-term and multi-factor effects.^2–4 Moreover, for many major diseases, early prediction and intervention are necessary to prevent and reduce the chance of damage to the mother–infant.⁵ In China’s medical industry, medical resources are tight and concentrated in some large-scale hospitals. How to make the patient’s visit convenient is also an urgent problem for medical institutions.

Traditional maternal and child care equipment is relatively large and cumbersome, and conventional monitoring is generally carried out in the hospital. Pregnant women need to travel back and forth between their home and the hospital, which is troublesome for pregnant women. In addition, the transmission of data through cables can make pregnant women nervous, and cable length restricts a pregnant woman’s actions, affecting the quality of monitoring. In summary, there are still some technical and practical limitations to traditional maternal and child care equipment.

Mobile healthcare, the delivery of healthcare via mobile communication devices, can optimize the traditional diagnosis and treatment model and improve the efficiency of the medical industry. It can also better optimize the allocation of hospital resources and improve the patient’s medical experience, and provide patients with better health management services.

The Yunban company built a “cloud + end” mobile medical system based on maternal and child care fields. They also built a large data cloud platform and researched and designed cloud platform–based portable medical equipment, which can be used for real-time detection of a variety of physical health indicators. Their system is capable of rapid analysis and real-time diagnosis of health data.

Related work

Filtering algorithm

In 1960, Kalman⁶ published his famous paper describing a recursive solution to the discrete-data linear filtering problem. Since that time, due in large part to advances in digital computing, the Kalman filter has been the subject of extensive research and application, particularly in the area of autonomous or assisted navigation.

The Kalman filter is a set of mathematical equations that provides an efficient computational (recursive) means to estimate the state of a process in a way that minimizes the mean of the squared error. The filter is very powerful in several aspects: it supports estimations of past, present, and even future states and can do so even when the precise nature of the modeled system is unknown.⁷

Estimation process

State prediction

The Kalman filter estimated the prior estimation of the state ${\hat{X}}_{k | k - 1}$ of the k-time by using the state ${\hat{X}}_{k - 1 | k - 1}$ of the last k − 1 step, which is based on the system state equation

{\hat{X}}_{k - 1 | k - 1} = A_{k} X_{k - 1} + B_{k} U_{k}

(1)

Priori estimates of covariance prediction

After we predicted the prior state of the system, we predicted the covariance matrix of the system state

P_{k | k - 1} = A_{k} P_{k - 1} + B_{k} U_{k}

(2)

$P_{k | k - 1}$ represents the a priori covariance of the system state, and $P_{k - 1 | k - 1}$ is the posteriori estimation covariance of the time step k − 1 of the system, which also means the estimated covariance of the system update at time step k − l.

Updating process

Kalman gain update

The Kalman filter is the key to the Kalman filter algorithm calculation, and it is optimal when the Kalman gain is effective, that is, when the minimum mean square error is obtained. We can obtain the Kalman gain as follows

K_{k} = P_{k | k - 1} H_{k}^{T} (H_{k} P_{k | k - 1} H_{k}^{T} + R_{k})^{- 1}

(3)

Status update

We updated the state of the system by combining the predictive value of the system transcendental state, the Kalman gain, and the system’s known measurement state values

{\hat{X}}_{k} = {\hat{X}}_{k | k - 1} + K_{k} (z_{k} - H {\hat{X}}_{k | k - 1})

(4)

Posteriori estimation covariance update

After getting the posterior state of the system, we updated the error covariance of the system at time step k

\begin{matrix} P_{k_{K}} = (I - K_{k} H_{k}) cov (X_{k} - {\hat{X}}_{k | k - 1}) (I - K_{k} H_{k}) \\ + K_{k} cov (V_{k}) K_{k}^{T} \end{matrix}

(5)

The Kalman filter recursively conditioned the current estimate of all of the past measurements, and then, in accordance with the above formula, became the best form of the system to estimate the current state and calculate the other new value.

Repair algorithm

Sparse representations are becoming an increasingly useful tool in the analysis of audio signals. Audio repair based on sparse representation refers to the use of sparse reconstruction in the overcomplete dictionary, using the data of the reliable part of the damaged audio, and then utilizes the sparse reconstruction method to recover the damaged part.

INcoherent K-singular value decomposition algorithm

Adler et al.⁸ first applied the sparse representation theory in image restoration to audio repair and proposed a sparse representation of the audio repair algorithm framework. In view of the shortcomings of fixed dictionaries, Mach and Ozdobinski⁹ used K-singular value decomposition (SVD) dictionary training methods to correct audio. The dictionary training method is essentially a solution to minimize formula (6)

min_{D, X} ∥ Y - DX ∥_{F}^{2} s . t . ∥ x_{i} ∥_{0} ⩽ S, i = 1, 2, \dots, K

(6)

The relevance of the dictionary is defined as the maximum absolute value of the inner product of two different atoms normalized, which reflects the maximum degree of similarity between the atoms in the dictionary. The mathematical formula is as follows

μ D = | 〈 \frac{d_{i}}{‖ d_{i} ‖} \frac{d_{j}}{‖ d_{j} ‖} 〉 | d_{i} d_{j} \notin D, i \neq j

(7)

Mailhé et al.¹⁰ proposed an INK-SVD algorithm based on the minimization of the matrices of the matrix dictionary learning method. Compared to the traditional dictionary training algorithm, the INK-SVD training algorithm added a low correlation constraint on equation (6)

min_{D, X} ∥ Y - DX ∥_{F}^{2} s . t . {\begin{matrix} ∥ x_{i} ∥_{0} ⩽ S, i = 1, 2, \dots, K \\ μ (D) ⩽ \bar{μ} \end{matrix}

(8)

OMP algorithm

Adler proposed audio repair, using a traditional orthogonal matching pursuit (OMP) algorithm to solve the sparse coefficient. Sparse representation, in mathematics, can be attributed to equation (9) optimization problem

X_{i} =_{x} ‖ x ‖ s . t . ‖ y_{i}^{r} - M_{i} D_{X} ‖ ⩽ ε

(9)

A common means to obtain approximate sparse solutions is to use matching pursuit (MP), OMP, and other greedy algorithms. There are two main steps in each of the MP and OMP’s iterations: (1) Selection of atoms: Select the atoms most relevant to the error from the dictionary and add them to the collection. (2) Update error: after selecting the atoms, recompute the error and continue the next iteration. The biggest difference between MP and OMP is that the OMP is normalized to the selected atom, and thus, it does not select repetitive atoms, and each time iteration makes the error smaller; it eventually gets the sparse approximation of the signal and has fast convergence. Conversely, the MP algorithm is slow to converge and can only generate suboptimal sparse representation after several iterations.

Yunban remote fetal monitoring project

Fetal heart rate (FHR) monitoring, used to detect hypoxia, fetal movement, contractions, and so on, gives important information about the health of the fetus during pregnancy.^11–15 However, there are still questions that must be addressed. (1) There is a lack of intelligent terminal equipment and cloud platform support. At present, the research of telemedicine platforms at home and abroad mainly focuses on the collection and transmission of data. However, there is relatively less focus on equipment portability, intelligent diagnosis of the terminal, and data analysis of the cloud. (2) There are few studies on the large data analysis of multi-source health data. The use of remote medical intelligent terminal processes will produce a lot of real-time detection data. How to use large data analysis technology, based on the effective integration of health data from a variety of sources, large-scale health data for individual users, and group-oriented analysis and processing, presently lacks research.

Project overview

The Yunban Remote Fetal Monitoring project, under the development of Internet Plus Mobile Medical, is a co-operation of production and research projects by the reporting company and participating company. This project focused on the subdivisions of pregnancy and child care, which is designed for maternal and infant parenting research and the development of mobile health medical large data platforms. The effective integration of resources through networking technology establish mobile medical health big data management platform for infant care. These include close contact with specialist doctors, medical institutions, and patients; medical devices; intelligent polymerization mobile user terminals, hospitals, and medical institutions with information systems; a social environment; and a variety of data resources. A schematic diagram of the overall project plan is shown in Figure 1.

Figure 1.

Schematic diagram of the overall project plan.

Remote fetal monitoring hardware equipment

The goal of the mother-to-child portable intelligent medical terminal is to remotely acquire basic human life data and conduct real-time analysis. It has GPRS, Bluetooth, WiFi, and TCP/IP networking functions and can connect to the health medical platform based on large data. The portable mobile intelligent terminal system topology is shown in Figure 2.

Figure 2.

Portable mobile intelligent terminal system topology.

The portable intelligent medical terminal system features include the following: (1) multi-function portable hardware design. Together portable mobile medical box’s features are integrated electrocardiogram (ECG), blood pressure, blood oxygen, and related processing circuits. This design can be applied to the needs of an ordinary home environment and it is easy to use. The system is divided into three parts. The first part detects the FHR, noninvasive blood pressure, ECG, and other test part; the second part conducts power conversion; and the third part deals with multipoint control unit (MCU) control and data transmission. (2) Adaptive life signal composite filter. Human life characteristics such as heart rate signals are voltage signals, generally between 0.5 and 5 mV, which is very weak compared to the outside of the power frequency interference. While untreated, life characteristics of the signal will be submerged in the noise, making serious interference for ECG signals diagnosis. In order to facilitate the signal for the late diagnosis, you must filter out the interfering ECG signal using the existing smooth filter, wavelet transform filter, infinite impulse response (IIR) filter, or finite impulse response (FIR) filter.

Data and methods

General information

We selected 160 cases of pregnant women in late pregnancy who were admitted to our hospital, Shenzhen Guangming New District Central Hospital, outpatient service from January 2016 to May 2017 as research subjects. They were randomly divided into an experimental group and control group, with 80 cases in each group. In the experimental group, the mother’s age was 22–40 years and the gestational age was 36–41 weeks. These patients underwent remote fetal monitoring and fetal movement counting. In the control group, the mothers were 22–38 years, and the gestational age was 37–41 weeks; these patients underwent fetal movement counting and routine FHR monitoring. The study was approved by the hospital medical ethics committee. The participating patients were willing to participate and signed informed consent. There was no significant difference (p > 0.05) between the two groups in age, pregnancy, and other aspects.

Improving the protection of the fetus in late pregnancy can effectively prevent maternal and child complications and further improve the quality of perinatal care. The current commonly used mode of monitoring fetal movement is simple, and for some women, the method cannot accurately reflect the situation of the fetus.^16,17 Remote fetal monitoring technology is a combination of fetal monitoring and telemedicine technology. The current commonly used method involves telephone network real-time fetal heart sounds and fetal movement information transmitted to the hospital computer central monitoring station; then, the doctor can monitor, analyze, and timely diagnose abnormal situations. This study presents an analysis of fetal monitoring in 80 cases of late pregnancy.

Experimental methods

Collected fetal heart signals contain a lot of noise after hardware processing, which has a great influence on experimental results. Thus, we used the Kalman filter noise reduction algorithm and repair algorithm, which we mentioned in section “Related work,” to ensure the accuracy of data acquisition.

Experimental group. We used the Doppler FHR monitor at a fixed time to listen to the fetal heart 3–4 times a day. When there were fetal heart or fetal movement abnormalities, we transmitted those data to the remote fetal care center immediately by phone. In accordance with the doctor’s guidance, we found the most prominent position of the FHR and kept contact with the Doppler fetal heartbeat and send the fetal signal to the hospital’s remote fetal monitoring center, while the time is generally 20–30 min. The patient used the phone’s “#” key when there was fetal movement during the whole monitoring process. The patient could request remote fetal monitoring at any time of the day until childbirth. If there were abnormal circumstances, the pregnant women immediately underwent bedside FHR monitoring and were placed in the hospital for observation and treatment. The monitoring report was analyzed by a specialist, who immediately informed the pregnant women of the results. During the monitoring process, other objective factors should be ruled out, like fetal sleep and the pregnant woman’s medication use, hunger, supine position, and so on. Control group. Pregnant women conducted their own fetal movement count at every 3 times a day for 1 h each time and at the same time paid attention on regular routine in the outpatient FHR monitoring (non-stress test).

Observation of indicators and judgments

NST monitoring abnormalities:¹⁸ (1) the baseline of the FHR was abnormal if more than 160 beats/min or less than 120 beats/min, (2) the reduction or disappearance of fetal heartrate–baseline variability, and (3) nonresponsive FHR was defined as an FHR without fetal acceleration or an acceleration rate of less than 15 times/min and a duration of less than 15 s or less than three times within 20 min.

Neonatal asphyxia classification:^19–21 We used the neonatal Apgar score method, scoring at birth or 5 min later. Scores of 1–3 indicated severe asphyxia, 4–7 indicated moderate asphyxia, and greater than 7 points indicated good neonatal condition.

Data processing

Using SPSS 17.0 statistical software for data processing, with (A ± S), we counted data using the χ² test and determined significance using a t test. The p < 0.05 was considered to be statistically significant.

Results

Comparison of NST abnormalities in newborns

In order to compare the NST abnormality of newborns, we found that the incidence of NST abnormality was 33.6% in the experimental group and 17.3% in the control group; the difference was statistically significant (p < 0.05). The results are shown in Table 1.

Table 1.

Comparison of neonatal group asphyxia.

Group	Cases	Good (n, %)	Moderate apnea (n, %)	Severe asphyxia (n, %)	Asphyxia rate (n, %)
Experimental	80	70 (87.5)	8 (1.0)	2 (0.3)	10 (12.5)*
Control	80	56 (70)	17 (21.2)	7 (8.8)	24 (30)

Compared with the control group, the difference was statistically significant: p < 0.05.

Comparison of neonatal asphyxia

Comparing the asphyxia situation of newborns, we found that the experimental group showed moderate asphyxia in eight cases and severe asphyxia in two cases. The incidence of neonatal asphyxia was 12.5%, which was significantly lower than in the control group (30%; p < 0.05). The results are shown in Table 2.

Table 2.

Comparison of abnormal neonatal NST in newborns.

Group	Cases	Fetal monitoring frequency (time)	NST exception (times)	Incidence of abnormality (%)
Experimental	80	232	78	33.6*
Control	80	81	14	17.3

NST: non-stress test.

Compared with the control group, the difference was statistically significant: p < 0.05.

Discussion

In late pregnancy, if the load of the fetus exceeds what the mother can bear, it will cause a variety of pregnancy-related complications, thus resulting in a significant decline of oxygen in the placenta. This, in turn, may lead to fetal distress or even stillbirth.^19–21 Pregnant women must always monitor the fetal heart, but because this can only be done in the hospital, pregnant women usually have to wait for several hours under various physical and mental circumstances. This also increases the economic burden on the mother. Remote fetal monitoring combines fetal monitoring and remote communication technology. Through this technology, pregnant women do not have to be hospitalized for monitoring, and medical staff can easily learn about the fetus’s intrauterine reserve, and whether there are complications, such as intrauterine hypoxia.

In this study, women in late pregnancy were selected as the study subjects. They were randomly divided into two groups. The experimental group was treated with remote fetal monitoring, while the control group was routinely hospitalized for traditional fetal monitoring. Upon follow-up, the NST abnormality was 33.6% in the experimental group and 17.3% in the control group. Complications can be detected through monitoring fetal movement and fetal heart. Traditional fetal movement counts can show the status of the fetus, but due to the influence of subjective factors, fetal movements cannot accurately reflect the fetal situation. In contrast, remote NST monitoring can better understand the fetal reserve capacity and can thus permit timely detection of abnormalities. With conventional NST and monitoring, these abnormalities cannot be found in time. Some scholars have suggested that the time gap of more than 2 days between fetal monitoring is likely to miss some abnormal fetal heartbeats, which has little value for the fetal prognosis predictions, so the daily or continuous monitoring is recommended. The results of this study showed that long-term fetal monitoring can be carried out at any time with remote monitoring. With NST monitoring and extended monitoring times, we can enhance the detection rate of abnormal NST and thus reduce adverse outcomes.

Changes in late-pregnancy fetal situations can lead to placental dysfunction, abnormal blood gas exchange between the mother and child, and chronic hypoxia. Once the risk factors worsen, the fetus would face severe hypoxia. Studies have found that high-risk obstetrics, such as hypertensive disorders and gestational diabetes, can lead to high maternal and neonatal mortality. Comparing the asphyxia situation of newborns, there were eight cases of moderate asphyxia in the experimental group and two cases of severe asphyxia. The incidence of neonatal asphyxia was 12.5%, which was significantly lower than the 30% in the control group. This suggests that fetal hypoxia can be timely and accurately diagnosed and treated through long-term remote monitoring, thus this can reduce the incidence of iatrogenic premature birth.

Remote fetal monitoring services will help to greatly reduce the fetal supervision pressure put on hospitals and reasonable distribution of monitoring needs, significantly reduce the labor demands on health care workers, and improve work efficiency. These services will also greatly alleviate the waiting time for pregnant women, ensure that they are more at ease, and provide them with social and economic benefits. With the awareness of perinatal health care and the improvement of living standards, high-risk pregnant women are eager to know the safety of their fetus at their own home. Establishment of the remote FHR monitoring network for monitoring at home can reduce worry of the fetus. Such a network can also shorten the hospitalization duration of pregnant women and reduce hospital costs and avoid unnecessary human intervention caused by premature delivery.

In summary, remote fetal monitoring for late pregnancy can detect NST abnormalities earlier and increase the detection rate of neonatal asphyxia. Therefore, it has great clinical value by reducing the incidence of adverse outcomes, improving the quality of obstetrics, and ensuring prenatal and postnatal care. It should be widely used in obstetric clinical monitoring.

Footnotes

Handling Editor: Pavel Stasa

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The authors gratefully acknowledge the contribution of the National Science Foundation of China (U1713212), (61572330), and (61602319) and the Technology Planning Project of Shenzhen City (JCYJ20170302143118519).

ORCID iD

Guokai Yan

References

Zeng

Hesketh

The effects of China’s universal two-child policy. Lancet 2016; 388(10054): 1930–1938.

Guelinckx

Devlieger

Beckers

et al . Maternal obesity: pregnancy complications, gestational weight gain and nutrition. Obes Rev 2010; 9(2): 140–150.

Clapés

Fernández

Suárez

Oxidative stress and birth defects in infants of women with pregestational diabetes. Medicc Rev 2013; 15(1): 37–40.

James

Jamison

MG.

Bleeding events and other complications during pregnancy and childbirth in women with von Willebrand disease. J Thromb Haemost 2010; 5(6): 1165–1169.

Milgrom

Holt

Early intervention to protect the mother-infant relationship following postnatal depression: study protocol for a randomised controlled trial. Trials 2014; 15(1): 385.

Kalman

RE.

A new approach to linear filtering and prediction problems. J Basic Eng: T ASME 1960; 82(D): 35–45.

Welch

Bishop

An introduction to the Kalman filter, vol. 8, issue 7. Chapel Hill, NC: University of North Carolina at Chapel Hill, 2010, pp.127–132.

Adler

Emiya

Jafari

et al . Audio inpainting. IEEE T Audio Speech 2012; 20(3): 922–932.

Mach

Ozdobinski

Optimizing dictionary learning parameters for solving audio inpainting problem. Int J Adv Telecommun Electrotech Signal Syst 2013; 2(1): 39–44.

10.

Mailhé

Barchiesi

Plumbley

. INK-SVD: learning incoherent dictionaries for sparse representations. In: IEEE international conference on acoustics, speech and signal processing, Kyoto, Japan, 25–30 March 2012, pp.3573–3576. New York: IEEE.

11.

Zhong

Wei

Zhang

LJ.

Predictive value of fetal heart rate (FHR) monitoring during the second stage of labor for fetal prognosis. J Clin Exp Med 2007; 3: 7–9.

12.

Deng

et al . Continuous-behavior and discrete-time combined control for linear induction motor-based urban rail transit. IEEE T Magn 2016; 52(7): 1–4.

13.

Deng

Luo

et al . A hybrid path planning method in unmanned air/ground vehicle (UAV/UGV) cooperative systems. IEEE T Veh Technol 2016; 65: 9585–9596.

14.

Deng

et al . Industrial internet: a survey on the enabling technologies, applications, and challenges. IEEE Commun Surv Tut 2017; 19: 1504–1526.

15.

Huang

Zhou

et al . Computation partitioning for mobile cloud computing in big data environment. IEEE T Ind Inform 2017; 13: 2009–2018.

16.

Pavlopoulos

Kyriacou

Berler

et al . A novel emergency telemedicine system based on wireless communication technology-AMBULANCE. IEEE T Inf Technol B 1998; 2(4): 261–267.

17.

You

et al . PSPEL: in silico prediction of self-interacting proteins from amino acids sequences using ensemble learning. IEEE/ACM T Comput Bi 2017; 14: 1165–1172.

18.

Ikemoto

Oishi

Onodera

et al . Retrospective analysis of chief complaints of patients with urogenital malignancies over the past decade at the Jikei University Hospital. Hinyokika Kiyo 2003; 49(2): 65–68.

19.

Lockshin

Kim

Laskin

et al . Prediction of adverse pregnancy outcome by the presence of lupus anticoagulant, but not anticardiolipin antibody, in patients with antiphospholipid antibodies. Arthritis Rheum 2012; 64(7): 2311–2318.

20.

Karlsson

Sporrong

Hillarp

et al . Prospective longitudinal study of thromboelastography and standard hemostatic laboratory tests in healthy women during normal pregnancy. Anesth Analg 2012; 115(4): 890–898.

21.

Kevane

McKenna

Walsh

et al . Haemorrhagic and thrombotic complications in pregnant women with acquired and congenital cardiac disease. J Perinat Med 2015; 43(2): 165–169.

Late pregnancy analysis with Yunban’s remote fetal monitoring system

Abstract

Keywords

Introduction

Related work

Filtering algorithm

Estimation process

State prediction

Priori estimates of covariance prediction

Updating process

Kalman gain update

Status update

Posteriori estimation covariance update

Repair algorithm

INcoherent K-singular value decomposition algorithm

OMP algorithm

Yunban remote fetal monitoring project

Project overview

Remote fetal monitoring hardware equipment

Data and methods

General information

Experimental methods

Observation of indicators and judgments

Data processing

Results

Comparison of NST abnormalities in newborns

Comparison of neonatal asphyxia

Discussion

Footnotes

Declaration of conflicting interests

Funding

ORCID iD

References