**Fuzzy Detection of Fetal Distress for Antenatal Monitoring in Pregnancy with Fetal Growth Restriction and Normal**

Igor V. Lakhno, Bertha Patricia Guzmán-Velázquez and José Alejandro Díaz-Méndez

Additional information is available at the end of the chapter

http://dx.doi.org/10.5772/intechopen.80223

#### Abstract

Monitoring of fetal cardiac activity is a well-known approach to the assessment of fetal health. The fetal heart rate can be measured using conventional cardiotocography (CTG). However, this method does not provide the beat-to-beat variability of the fetal heart rate because of the averaging nature of the autocorrelation function that is used to estimate the heart rate from a set of heart beats enclosed in the autocorrelation function window. Therefore, CTG presents important limitations for fetal arrhythmia diagnosis. CTG has a high rate of false positives and poor inter- and intra-observer reliability, such that fetal status and the perinatal outcome cannot be predicted reliably. Non-invasive fetal electrocardiography (NI-FECG) is a promising low-cost and non-invasive continuous fetal monitoring alternative. However, there is little that has been published to date on the clinical usability of NI-FECG. The chapter will include data on the accurate diagnosing of fetal distress based on heart rate variability (HRV). A fuzzy logic inference system was designed based on a set of fetal descriptors selected from the HRV responses, as evident descriptors of fetal well-being, to increase the sensitivity and specificity of detection. This approach is found to be rather prospective for the subsequent clinical implementation.

Keywords: fetal non-invasive electrocardiography, fetal heart rate variability, fetal distress

#### 1. Introduction

Electronic fetal monitoring is an important part of the prenatal surveillance system that contributes to the best perinatal outcome. Its objective is the correct evaluation of fetal well-being.

© 2018 The Author(s). Licensee IntechOpen. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

However, the lack of precision of several methods that monitor the fetal well-being is well known [1].

Cardiotocography (CTG) methods have been standard, despite the lack of evidence that it reduces the adverse sequelae of neurodevelopment, including neonatal hypoxic-ischemic encephalopathy and cerebral palsy. This method has a high rate of false positives, and poor inter- and intraobserver reliability [2–4], such that fetal status and the perinatal outcome cannot be predicted reliably.

CTG is a widely available method for fetal development research, based on cardiac rhythm reactivity to fetal intrauterine motor activity in the prenatal period. Doppler ultrasound is the most obvious technological approach for monitoring fetal well-being. However CTG-based techniques require prolonged ultrasonic monitoring.

CTG demonstrates the response of the sinus node to the continuous interaction of the sympathetic and parasympathetic tones of the autonomic nervous system [5, 6]. Autonomic control of fetal cardiac rhythm could be investigated by fetal heart rate variability (HRV). HRV captures the impact of central and peripheral circuits on regulation in hemodynamics [7]. The recording of primary bioelectrical processes in the sinus node can be assumed to be a more valuable technique than the mechanical detection of cardiac cycles used in CTG [1].

The fetal HRV parameters exhibit a wide range, even under normal conditions. The peculiarities of the fetal neurobehavioral response in the active and sleepy periods may complicate the interpretation of the conventional CTG tracing, and increase the level of cesarean interventions [8–10].

Recent research has explored different biochemical and biophysical markers, as well as the correlation between maternal-fetal hemodynamic processes, to better understand the complex processes involved in the loss of fetal well-being [11–20].

The objective of this study is the design of a fuzzy inference system based on a set of fetal descriptors, selected from the CTG and HRV responses, as evident markers of fetal well-being, to increase the sensitivity and specificity in evaluation of fetal distress.

#### 2. Analysis and selection of descriptors

For the development of this study, records of 49 pregnant women were used. These were taken in the Department of Maternal and Fetal Medicine of Kharkiv municipal perinatal center. These records were divided into four groups: Group I composed of healthy pregnant women without loss of fetal well-being, group II of healthy pregnant women with loss of fetal wellbeing, group III of pregnant women of high-risk type III without loss of fetal well-being, and group IV of pregnant women of high-risk type III with loss of fetal well-being. NI-FECG tracing was obtained from the maternal abdominal wall using the Cardiolab Babycard equipment (Scientific and research center "KhAI Medica," Ukraine) [1, 11, 12]. The sampling rate was 1000 Hz. For all reported cases, the study protocol was approved by the Bioethics Committee of the Kharkiv Medical Academy of Postgraduate Education (registration number 0105 U002865). For training purposes of fuzzy inference system, the 49 records were divided into windows of 2 minutes to obtain 296 datasets of HRV and CTG parameters.

In order to select the best descriptors to design the fetal well-being inference system, an observation was made using ROC curves and Spearman correlation of the different fetal HRV and CTG parameters used in [12] and shown as the best correlated with Apgar Score 1. These parameters are shown in Table 1.

The specificity (Sp) and the sensitivity (Se) of the parameters concerning the fetal well-being were obtained from the ROC analysis. Sp and Se are given by Eqs. (1) and (2), respectively:

$$S\_p = \frac{VN}{VN + FP} \tag{1}$$

where VN are the true negatives and FP are the false positives.


Table 1. HRV and CTG parameters.

$$S\_t = \frac{VP}{VP + FN} \tag{2}$$

where VP are the true positives and FN corresponds to the false negatives.

Figure 1. ROC curves for HRV variables. (a) SDNN AUC = 0.8653, (b) RMSSD AUC = 0.8922, (c) pNN50 AUC = 0.7982, (d) SI AUC = 0.9956, (e) AMo AUC = 0.917, (f) TP AUC = 0.8614, (g) VLF AUC = 0.8514, (h) LF AUC = 0.898, and (i) HF AUC = 0.8976.

Spearman's correlation (r) between bio-signal parameters and well-being fetus state is given by:

$$\text{res} = 1 - \frac{6\sum\_{i=1}^{n} d\_i^2}{n\left(n^2 - 1\right)}\tag{3}$$

where di is the ranges of values for i-parameter and the clinical diagnosis.

Figures 1 and 2 show the ROC curves for the HRV and CTG fetal parameters. As can be seen, the highest AUC is obtained for AMo = 0.9170 and SI = 0.9956 for HRV and, ACC = 0.9974, LTV = 0.9950, STV = 0.9972 and LOWVAR = 0.9922 for CTG. The smallest area was obtained for PNN50 = 0.7982 and DES = 0.6071 for HRV and CTG, respectively.

Table 2 shows the results of the sensitivity, specificity, and Spearman's correlation for HRV parameters which are also shown in Figure 3. SDNN, which measures the general variability of the neurovegetative system, showed a high Sp = 1 and a Spearman's correlation of �0.6352,

Figure 2. ROC curves for CTG parameters: (a) STV = 0.9772, (b) LTV = 0.9950, (c) ACC = 0.9974, (d) DES = 0.6071, (e) LowVar = 0.9922, and (f) HighVar = 0.8353.


Table 2. AUC, sensitivity, specificity, and Spearman's correlation of fetal HRV parameters.

Figure 3. AUC, sensitivity, specificity, and Spearman's correlation of HRV parameters in the study population.

however showed a low value of Se = 0.7765. RMSSD, which is related to high-frequency components, like HF are below 0.90 in the values of both Se = 0.7765 and Sp = 0.8235. TP and LF, although they have a high specificity, 0.9647 and 0.9882 respectively, have a low sensitivity of 0.7882 and 0.7765, respectively. VLF although it has a high specificity Sp = 1 and a Spearman's correlation of �0.6089, its sensitivity is low, Se = 0.7882; HF also presents a low sensitivity of 0.7765. Pnn50 had the lowest sensitivity and specificity Se = 0.7765 and Sp = 0.7412. SI presented Se = 0.9882 and Sp = 1; and AMo showed Se = 0.9882 and Sp = 0.8112. The Spearman's correlation for SI and AMo are, respectively, 0.8585 and 0.7243, which is consistent since SI and AMo are independent of the steady state of the fetus and can be considered evident markers of fetal well-being [12].

In the same way in Table 3 are shown AUC, Se, Sp, and r for CTG parameters, these are shown also in Figure 4. As can be seen, the highest values were obtained for: STV with a Se = 0.9765, Sp = 1, and r-Spearman = �0.8271. LTV with a Se = 0.9882, Sp = 1, and r-Spearman = �0.8579. Fuzzy Detection of Fetal Distress for Antenatal Monitoring in Pregnancy with Fetal Growth Restriction… 15 http://dx.doi.org/10.5772/intechopen.80223


Table 3. AUC, sensitivity, specificity, and spearman's correlation of fetal CTG parameters.

Figure 4. AUC, sensitivity, specificity, and Spearman's correlation of CTG parameters in the study population.

The accelerations (ACC) with Se = 0.9882, Sp = 0.9882, and r-Spearman = �0.8826. LOWVAR with a Se = 0.9765, Sp = 1, and r-Spearman = 0.9218. Although HIGHVAR shows a high Se = 1, its Sp is low of 0.6706. The evaluation of short-term variations (STV) and long-term variations (LTV) allow that can be used as markers of fetal compromise.

#### 3. Fuzzy inference system design

The inference system of the fetal well-being state was designed with fuzzy logic, Mamdani-type, 4 inputs (SI, AMo, STV, and LTV), 1 output (status of fetal well-being), and 16 fuzzy rules. The block diagram is shown in Figure 5. The fuzzy logic design allows us to take advantage of the linguistic interpretation capacity in complex problems, when there is no simple solution model or a precise mathematical model, such as the detection of loss of fetal well-being. The ranges of selected descriptors are shown in Table 4, and were used as the basis for the design of the fuzzy membership functions.

The input variables for fuzzy inference system are SI and AMo for HRV and STV and LTV for CTG. Figure 6 shows the input membership functions for these variables.

Figure 5. Fuzzy inference system for loss of fetal well-being detection.

Fuzzy Detection of Fetal Distress for Antenatal Monitoring in Pregnancy with Fetal Growth Restriction… 17 http://dx.doi.org/10.5772/intechopen.80223


Table 4. Ranges of values of the fetal status descriptors.

Figure 6. Membership functions for input descriptors: (a) SI, (b) AMo, (c) LTV, and (d) STV.

Figure 7. Membership functions for fuzzy system output sets.

The fuzzy output is shown in Figure 7, and this is defined by two trapezoidal membership functions, representing normal (N) or distress (D) fetal status, and by a triangular function in which the diagnosis is indeterminate by fuzzy system. The "Normal" output is in the range from 0 to 0.4; for the output "Distress," the rank is of 0.6 to 1.0; and if the output is between 0.4 and 0.6, it is classified as "indeterminate."

The fuzzy knowledge base is shown in Table 5. In order to increase the presumption of fetal well-being detection, the inference would have to be "Normal," if and only if, the membership


Table 5. Fuzzy knowledge base for well-being fetal status.

of the fuzzy sets in the four input descriptors belongs to "Normal" set. On the other hand, the output will be "distress" if the membership in at least one variable of HRV and CTG belong to the input set related to fetal distress. Fuzzy inference can be indeterminate if membership of fuzzy sets in three inputs belongs to sets related to normal fetal status. The "Normal" output is in the range of 0–0.4; "Distress" output range is from 0.6 to 1.0, and the output range from 0.4 to 0.6 will be classified as "Indeterminate."

The fuzzy rules can be written as:
