Noninvasive Fetal Heart Rate Monitoring : Validation of Phonocardiography-Based Fiber-Optic Sensing and Adaptive Filtering Using the NLMS Algorithm

Here we present the evaluation results of our novel noninvasive phonocardiographic-based fiberoptic sensor for fetal Heart Rate (fHR) detection using adaptive filtering and the NLMS Algorithm. The sensor uses two interferometric probes encapsulated inside a PolyDiMethylSiloxane (PDMS) polymer. Based on real data acquired from pregnant women in a suitable research laboratory environment, once they had given their written informed consents, we created a simplified dynamic signal model of the distribution of maternal and fetal heart sounds inside the maternal body. Building upon this signal model, we verified the functionality of our novel fiber-optic sensor and its associated adaptive filtering system using the NLMS Algorithm. The main reason why we chose this technology to develop our system was that it allows monitoring the fHR without exposing the fetus to any external energies or radiation (in contrast to the ultrasound-based Cardiotocography Method). We used objective criteria such as: Signal to Noise Ratios: SNRin, SNRout and Percentage Root-mean-square Difference (PRD) for our evaluations.


Introduction
In this article, we report on the evaluation of a noninvasive method for fetal Heart Rate (fHR) detection and monitoring during gestation, labor, and delivery based on fetal PhonoCardioGraphy (fPCG).Our proposed method relies on the combined capabilities of fiberoptic sensing and adaptive filtering (implementing the Normalized Least Mean Square -NLMS -Algorithm).In our recent work reported elsewhere [1] and [2], we developed an adaptive system, which enabled us to measure the fetal Heart Rate (fHR) by means of fPCG signal peak detection using the maternal abdominal PhonoCardioGrams (aPCGs).We observed that diagnostic-quality fPCG signals required for accurate fHR detection are contaminated by an unwanted maternal component (the mPCG signals) in addition to other technical and biological interferences.We showed that as the spectral contents of the fPCG and mPCG signals overlap in the frequency domain, common filtering methods such as signal subtraction, linear filtering, and others are ineffective in extracting reliable fHR information and therefore cannot be used.
The well-established conventional Phonocardiography is based on the scanning of acoustic signals by means of a microphone placed on the thorax.As for fetal Phonocardiography, the microphone is placed on the maternal abdomen [12], [13] and [14].
Our solution described here is based on the scanning of acoustic signals by means of two Mach-Zehnder interferometric fiber-optic probes.The advantages of these interferometers are their immunity to Electro-Magnetic Interferences (EMI), and their ability to measure any changes in the optical path length (such as the core refraction index, fiber length and the wavelength used).Therefore, the smallest measurable frequency due to any phenomena resulting in the change of the above-mentioned physical properties is theoretically unlimited [15] and [16].To perform our system evaluations, we needed to use synthetic data.For generating suitable synthetic signals, we conducted a set of measurements on pregnant women in a suitable research laboratory environment after obtaining their written consents.We then created a simplified dynamic signal model for the distribution of maternal and fetal heart sounds inside the maternal body.Based upon this signal model, we generated synthetic data with properties as close as possible to the real data.The necessity to use synthetic data at this stage of our research was further justified by considering the fact that our patent-pending interferometric sensors have yet to be legislatively approved for clinical testing on pregnant women.It is important to emphasize that legislative regulations for use of new technology on pregnant women are extremely strict (as an unborn fetus is critically sensitive to external energies such as mechanical pressure, electromagnetic radiation, change in temperature, and so on).
In current clinical practice, clinicians use either ultrasound-based methods such as CardioTocoGraphy (CTG), which measures the fetal heart rate along with maternal uterine contractions, or fetal Echocardiography (fECHO) to diagnose fetal congenital heart defects from the 20 th to the 23 rd week of pregnancy [17] and [18].These sophisticated technologies are now integral parts of routine modern obstetrics.It is important to emphasize that the CTG technology has helped clinicians reduce the mortality rate of newborn babies during delivery.In spite of this considerable impact, it is generally recognized that this technology has some disadvantages such as high sensitivity to noise caused by maternal movements and the need to frequently reposition the ultrasound transducers.Also, this method is not suitable for long-term continuous fetal heart rate monitoring due to the potentially harmful influence of ultrasonic radiation on the fetus.Our method and system, once statistically and clinically proven and validated, offer a number of advantages (in contrast to the currently used ultrasoundbased CTG and other conventional methods), including their applicability to continuous long-term fHR monitoring without exposing the fetus to any radiation as well as their compatibility with Magnetic Resonance Imaging (MRI) environments.The continuous long-term monitoring capability of our system is highly desirable, especially in those cases in which the pregnant woman faces a dangerous situation (such as after an accident), and it becomes absolutely essential to perform a time consuming MRI examination to ensure that the unborn child is intact and safe.The other specific advantage of our technology is that it can be used in water deliveries.

Fetal Phonocardiography
Fetal PhonoCardioGraphy (fPCG) was discovered during the 17 th century by Kergardec, Marsac, and Kennedy [19].Although fPCG was discovered a very long time ago, interest in this research area has only grown over the past few years.This figure shows the number of peer-reviewed articles that appear in the Science Direct, the Institute of Electrical and Electronics Engineers (IEEE) and the National Institute of Health (PubMed) databases.
The PCG signal is composed of two main acoustic components (the first heart sound S1 and second heart sound S2), see Fig. 1, and two additional heart sounds (S3 and S4).S1 is systolic and is connected with the closure of bicuspid and tricuspid valves at the beginning of ventricular contraction.S2 is diastolic and is produced by the closure of semilunar valves.The third heart sound (S3) is pro-diastolic and appears when a valve muscle quivers during the fast phase of blood flow into the valve.The fourth heart sound (S4) is presystolic and is a sign of the quivering of valve muscle during systole in the atrium.The last two mentioned heart sounds (S3 and S4) are not common for adults, and their presence is a sign of cardiac insufficiency [20].
Fig. 1: Basic components of PCG signals.

PCG-Based Fibre-Optic Sensor
Our fiber-optic sensor is encapsulated inside polydimethylsiloxane [21], [22] and [23] and is comprised of two Mach-Zehnder interferometric components formed by 1×2 and 3×3 power couplers with an even split ratio; see Fig. 2. The reference fiber is stored in a stable environment.The output beams are recombined at a second 3×3 coupler.The output signal is detected by photodetectors.The resultant optical intensity after 3×3 coupler can be described by the following Eq.( 1).
where n represents the coupler output index with a value of 1, 2 or 3.The symbol A n represents the mean value of optical intensity (DC component).Symbol B n represents the optical intensity variation amplitude depending on fringe visibility, φ(t) represents the signal of interest, and φ drif t (t) is a quasi-static phase shift due to coupler properties.For the extraction of the proper signal, it is necessary to use a demodulation algorithm [24].

Implementation of the Adaptive NLMS Algorithm
The measurands sensed by our interferometric sensors generated the fetal heart rate information, which was then fed into an adaptive stochastic system using the Root Mean Square Error (RMSE) criterion.This stochastic approach required a large number of measurements to produce powerful statistics.This consideration led to the utilization of the Normalized Least Mean Square (NLMS) Algorithm, which is a representative of basic stochastic gradient-based adaptation methods; see articles [25] and [26].
The Normalized Least Mean Square (NLMS) Algorithm is a variant of the Least Mean Square Algorithm.The former is able to accelerate the convergence speed with a reasonable computational cost and selects a normalized step-size µ n , which results in both a stable and fast converging adaptation algorithm, see [27] and [28].Implementation of the NLMS Algorithm can be summarized as follows: The step-size µ n can be described as follows Eq. (2).

Results
Our measurement system comprised of a novel fiberoptic sensor and its associated adaptive filtering system for fetal Heart Rate (fHR) monitoring is shown in Fig. 3.The adaptive system consists of two measurement sensors (FC/APC type) which were placed on the chest and abdomen, optical interrogator and DSP (Digital Signal Processing) unit for the recording, amplification, digitalization, demodulation and filtering the measured signals.Optical interrogator consists of DFB (Distributed Feedback Laser) laser with wavelength 1549.5 nm and output power of 3 mW and three InGaAs Amplified Photodetectors (Indium Gallium Arsenide).Signal was digitalized by National Instruments card NI-USB 6210 with the sampling frequency of 250 kHz and analyzed by software application written in the LabView (2015, National Instruments, Austin, Texas, USA) [29] and [30].
Fig. 3: Basic scheme of our sensor and its associated NLMS adaptive system for fHR monitoring.
Measurements (Fig. 4) were performed in a suitable research laboratory environment on 8 pregnant women (GA = 36-42 weeks) after obtaining their written informed consents.The test subjects were between the age of 21 and 37, their weight was between 57 kg and 103 kg, and their height was between 156 cm and 196 cm.Based on the obtained results we can state that no significant differences were found in the quality of the collected data based on the subjects' age, weight, and height.Using real data, we created a simplified dynamic model of sound distribution in the human body to generate suitable synthetic signals such as: ST (signals from sensors placed on the chest) and SA signals (from sensors placed on the abdomen).Our PCG signal model was inspired by contributions made by AL-MASI et al. [31] and [32], who devoted considerable efforts to generating synthetic PCG signals.In addition, we greatly benefited from our own research in generating realistic synthetic physiological and pathological fECG signals [33] and [34] in order to evaluate the performance of our system.
Figure 5 shows an ideal mPCG signal after removing the mother's breathing artifacts (using a Butterworth second-order band-pass filter with corner frequencies: f L = 10 Hz, and f H = 400 Hz, respectively).This signal served as a reference input for our adaptive system running the NLMS Algorithm.The filtered results enabled us to determine the mHR (by performing mPCG signal peak detection).Maternal first and second heart sounds are denoted as mS 1 and mS 2 , respectively, in Fig. 5. Figure 6 shows an ideal fPCG waveform after preprocessing the maternal signal.We need to emphasize here that the first fetal heart sounds (fS 1 ) result from the closing of the fetal tricuspid and mitral valves and the second fetal heart sounds (fS 2 ) are produced by the closure of the fetal pulmonic and aortic valves.nent has been significantly reduced.This figure clearly shows that the elimination of the maternal component is not ideal; nevertheless, this component is reduced well under the level of fPCG signals.Using the filtered signal, we can use conventional techniques [35], [36] and [37] to determine the fHR information from the fPCG signals.

Time (s)
Table 1 summarizes our experimental results.The performance of our adaptive system using the NLMS Algorithm was evaluated by finding the differences between input (SNR in ) and output (SNR out ) values as well as the objective measure known as the Percentage Root-mean-square Difference (PRD) [38].The SNR in value can be calculated by using the following Eq.( 4): where sig usef (n) is a desired signal (modelled reference course of S T ) and sig noise (n) is a noise signal (mPCG is measured up in the abdominal part -S A ).
The SN R OU T value can be calculated by using the following equation Eq. ( 5): where sig pre (n) represents a predicted (estimated) signal, or more precisely, the output from the proposed NLMS adaptive system and sig des (n) represents the desired signal.
One way to quantify the difference between the reference and the output signal: sig pre (n) is by using the PRD as given by equation Eq. ( 6) below:

Conclusion
In this article we focused on the validation of our novel patent-pending interferometric PPG-based sensor and its associated adaptive filtering system using the NLMS Algorithm for effective processing of aPCG signals to extract fPCG signals and fHR information.In the evaluations of the signal filtering quality of our system, we used objective parameters such as SNR and PRD.
The main reason why we chose the fiber-optic technology to develop our system was that it enables fHR monitoring without exposing the fetus to any radiation (in contrast to the ultrasound-based CTG method).Our innovative system offers a number of advantages including applicability to continuous long-term fHR monitoring without exposing the fetus to any radiation as well as compatibility with Magnetic Resonance Imaging (MRI) environments.The long-term monitoring capacity of our system is highly desirable, especially in those cases when the pregnant woman faces a dangerous situation (such as after an accident), and it becomes absolutely necessary to perform a time consuming MRI examination to ensure that the unborn fetus is intact and safe.The other specific advantage of our technology is that it can be used in water deliveries.
In our future research, we intend to use data from clinical practice to investigate a variety of challenging research topics such as the influence of sensor placement, fetal position and gestational age on aPCG signal filtering, fPCG signal extraction, and fHR monitoring.
Telecommunications.Two years later, he received his Master's degree in the field of Telecommunications in the same workplace.He is currently employee (science and research assistant) and a Ph.D. student of Department of Telecommunications at VSB-Technical University of Ostrava.He works in the field of fiber optic sensor systems.

Fig. 4 :
Fig. 4: An example of real data acquisition from a volunteer subject.

Fig. 5 :
Fig. 5: The reference synthetic mPCG signal based on real measurements made from thoracic (S T ) sensors.

Fig. 6 :
Fig. 6: The reference ideal synthetic fPCG signals based on real measurements from abdominal (S A ) sensors.

Figure 7
Figure 7 shows an example of the primary abdominal PCG (aPCG) synthetic input signal measured by the abdominal sensor.The aPCG signal (made up of the fPCG and mPCG components) is applied to the adaptive NLMS Algorithm.For determination of the fetal Heart Rate (fHR), it is necessary to detect fS1 components in the composite aPCG signals, which is a difficult task without advanced signal processing.

Figure 8 Fig. 7 :Fig. 8 :
Figure 8  shows an example of the output from our adaptive system using the NLMS Algorithm.Based on these results we can observe that: the mPCG compo- Tab. 1: Statistical results of the tested NLMS Algorithm.
FAJKUS was born in 1987 in Ostrava.In 2009 he received a Bachelor's degree from VSB-Technical University of Ostrava, Faculty of Electrical Engineering and Computer Science, Department of Telecommunications.Two years later, he received a Master's degree in the field of Telecommunications in the same workplace.He is currently employee and a Ph.D. student of Department of Telecommunications at VSB-Technical University of Ostrava.He works in the field of optical communications and fiber optic sensor systems.Stanislav KEPAK was born in 1987 in Ostrava.In 2011 he received Master's degree in the feld of Telecommunications.He is currently Ph.D. student and he works in the feld of optical communications and fiber optic sensor systems.Radana KAHANKOVA was born in 1991 in Opava, Czech Republic.She received her Bachelors's degree at the VSB-Technical University of Ostrava, the Department of Cybernetics and Biomedical Engineering in 2014.Two years later at the same department, she received her Master's degree in the field of Biomedical Engineering.She is currently pursuing her Ph.D in Technical Cybernetics.Her current research is focused on improving the quality of electronic fetal monitoring.Petr JANKU was born on 30 th May 1969 in Brno, Czech Republic.He graduated from the Medical Faculty Masaryk University in Brno.He works in the Department of Obstetrics and Gynecology of the University Hospital Brno as a Deputy Head responsible for Obstetrics.His research work concentrates on monitoring of fetuses.Vladimir VASINEK was born in Ostrava.In 1980 he graduated in Physics, specialization in Optoelectronics, from the Science Faculty of Palacky University.He was awarded the title of RNDr. at the Science Faculty of Palacky University in the field of Applied Electronics.The scientific degree of Ph.D. was conferred upon him in the branch of Quantum Electronics and Optics in 1989.He became an associate professor in 1994 in the branch of Applied Physics.He has been a professor of Electronics and Communication Science since 2007.He pursues this branch at the Department of Telecommunications at VSB-Technical University of Ostrava.His research work is dedicated to optical communications, optical fibers, optoelectronics, optical measurements, optical networks projecting, fiber optic sensors, MW access networks.He is a member of many societies: OSA, SPIE, EOS, Czech Photonics Society; he is a chairman of the Ph.D. board at the VSB-Technical University of Ostrava.He is also a member of habitation boards and the boards appointing to professorship.Homer NAZERAN holds B.Sc., M.Sc.and Ph.D. degrees in Electrical (Honors), Clinical and Biomedical Engineering from UT Austin, Case Western Reserve and University of Texas Southwestern Medical Center (UTSWM) at Dallas/UTA, respectively.He has close to 3 decades of experience in industry and academia and has practiced and taught biomedical engineering in the Middle East, Europe, Australia and USA.In Australia, with Professor Andrew Downing he co-founded the School of Engineering at the Flinders University of South Australia, introduced and established the electrical and electronics and biomedical engineering degree programs (1991 to 2001).He returned to the University of Texas at Arlington as a visiting professor in 1997 and 2001.He joined UTEP in 2002 to create and establish biomedical engineering degree programs at the Department of Electrical and Computer Engineering.His research interests are in the areas of computer modeling of physiological systems, intelligent biomedical instrumentation and biomedical signal processing as applied to chronic health conditions and telemedicine.He has more than 150 journal and conference articles in his research areas published in IEEE Engineering in Medicine and Biology Society (EMBS) and other flagship international conference proceedings.He is a reviewer for several national and international journals in his related fields including IEEE Transactions on Biomedical Engineering, Medical and Biological Engineering and Computing, Biomedical Engineering Online and others.His teaching interests are in electronics, biomedical instrumentation, physiological systems, and biomedical signal processing.He is also interested in development of novel teaching methods, lifelong learning and critical thinking habits in the classroom and interdisciplinary education based on application of nonlinear dynamics systems (complexity) theory.His research, teaching and professional activities have been supported by NIH, NSF, and DOE among others.Radek MARTINEK was born in 1984 in Czech Republic.In 2009 he received Master's degree in Information and Communication Technology from VSB-Technical University of Ostrava.Since 2012 he worked here as a research fellow.In 2014 he successfully defended his dissertation thesis titled "The use of complex adaptive methods of signal processing for refining the diagnostic quality of the abdominal fetal electrocardiogram".He works as an associate professor at VSB-Technical University of Ostrava since 2017.