Identification of patients with preeclampsia from normal subjects using wavelet-based spectral analysis of heart rate variability

Abstract

BACKGROUND:

The spectral analysis of the heart rate variability (HRV) shows a decrease in the power of the high frequency (HF) component in preeclamptic pregnancy compared with normal pregnancy; such a decrease is associated with an increase in the low frequency (LF) and the very low frequency (VLF) power. The physiological interpretation is that preeclamptic pregnancy is associated with a facilitation of sympathetic regulation and an attenuation of parasympathetic influence of HR compared with non-pregnancy and normal pregnancy.

OBJECTIVE:

To use an efficient nased on spectral analysis non-invasive technique to identify preeclamptic pregnant subjects from normal pregnant in Oman.

METHODS:

The soft-decision wavelet-based technique is implemented to find the power of the HRV bands in high resolution manner compared to the classical fast Fourier Transform method. Data was obtained from 20 preeclamptic pregnant subjects and 20 normal pregnant controls of the same pregnancy duration, obtained from Nizwa and Sultan Qaboos University hospitals in Oman.

RESULTS:

The soft-decision wavelet method succeeds to identify patients from normal pregnant with specificity, sensitivity and accuracy of 90%, 80% and 85%, respectively, compared to the FFT which results in 75% specificity, sensitivity and accuracy.

CONCLUSION:

The LF power obtained by Soft-decision wavelet decomposition is shown to be a successful feature for identification of preeclampsia.

Keywords

Preeclampsia normal pregnant identification HRV soft-decision wavelet spectral analysis

1. Introduction

In Muscat, the capital of Oman, a study showed a preeclampsia prevalence of 7.5% of all pregnancies [1]. Preeclampsia affects 10% of all pregnancies worldwide and causes substantial maternal and fetal morbidity and mortality [2].

The pathophysiological mechanisms underlying preeclampsia are not completely studied and understood. Impairment of the autonomic nervous system functions may be the cause of preeclampsia [3]. There is still a debate regarding whether preeclampsia is associated with disturbances in the sympathetic and parasympathetic functions of the autonomic nervous system [4].

Autonomic function can be assessed invasively by microneuroangiography and using several noninvasive laboratory techniques: Valsalva maneuver, the deep breathing test, orthostatic tolerance, ice immersion and handgrip dynamometry [5].

The information obtained from evaluation of autonomic activity using spectral analysis of heart rate variability (HRV), are summarized as [6]:

The very low-frequency (VLF) power-spectral density (PSD) components, between 0.0033 Hz and 0.04 Hz, which are possibly related to long-term regularity mechanisms (e.g., the rennin-angiotensin system, the thermoregularity peripheral blood flow adjustment).

The low-frequency (LF) components, between 0.04 Hz and 0.15 Hz, linked to sympathetic modulation, but also including some parasympathetic influence. The high-frequency components, from 0.15 Hz to 0.4 Hz, which reflect parasympathetic (vagal) activity.

The frequency-domain analysis of HRV has shown a decrease in the power of the high frequency (HF) component in preeclamptic pregnancy compared with normal pregnancy; such a decrease is associated with an increase in the LF/HF ratio of the HRV [2]. The physiological interpretation for that was normal pregnancy is associated with a facilitation of sympathetic regulation and an attenuation of parasympathetic influence of heart rate compared with non-pregnancy. Such alterations are influenced more in preeclamptic pregnancy compared to normal pregnancy [2]. Frequency domain analysis showed sympathetic dominance and vagal withdrawal in pre-eclamptic pregnant women [7].

The classical Fast Fourier transform (FFT) spectral analysis was used to show whether patients with preeclamptic pregnancy have a reduction in the power of the HF band of HRV compared with subjects with normal pregnancy [8]. The soft-decision wavelet-based technique is also implemented to scan the HF band to find which part of it is associated with preeclampsia. Data was obtained from the antenatal clinic of Nizwa hospital in Nizwa, Oman. Eight preeclamptic pregnant subjects and nine normal pregnant controls of the same pregnancy duration were used.

Another study is conducted in Oman on larger size of data [9]. The results indicate that the HF power of the HRV spectrum decreases in preeclamptic pregnancy compared with normal pregnancy.

The study in [10] showed considerable change in HRV and heart rate during pregnancy. Hence, we can conclude that it is important to understand the HRV or cardiovascularautonomic nervous system in normal pregnancy before being able to say whether there is any failure of its regulation in complicated pregnancy.

The inhibition of resting parasympathetic activity or vagal blockage and an increment of the sympathetic modulation during the 3rd third trimester of gestation in pregnancy as compared to their 1st trimester and healthy non-pregnant subjects were investigated [11].

Further evidence is added for the dominant cardiac sympathetic modulations on patients with preeclampsia, probably parasympathetic withdrawal in this group [12]. However, the higher VLF and LF readings achieved by preeclamptic women compared with the controls are unexpected in the view that augmented sympathetic modulations usually depresses all HRV parameters including these two measures.

The main objective of this paper is to develop a noninvasive classification method based on HRV with the great advantages over invasive microneuroangiography as it has minimum risk to the mother and the fetus. The classification method is to use the 40 subjects used in [9], in a complete identification system having 20 subjects as trial group and the other 20 subjects as test group.

The soft-decision power spectral estimation technique based on sub-band decomposition and wavelet-decomposition was implemented successfully in many applications such as identification of patients with obstructive sleep apnea and congestive heart failure and identification of parkinsonian tremor from essential tremor using accelerometer and surface EMG signals [13, 14, 15, 16, 17, 18]. This advanced spectral analysis method will be used in this work also.

The organization of the paper is as follows:

In Section 2, the data used in the work is described. In Section 2 also, the methods implemented in the work are investigated. Section 3 contains the main results of identifying preeclampsia from normal pregnancy using the soft-decision wavelet-decomposition and FFT. Section 4 includes conclusions of the work.

2. Data acquisition

2.1 Subjects

The data used in this study were collected from two hospitals in Oman. HRV of 17 subjects: 9 with normal pregnancy and 8 preeclamptic subjects, were obtained from the antenatal clinic of Nizwa hospital in Nizwa, Oman. HRV daata of another 23 subjects: 11 with normal pregnancy and 12 with preeclampsia pregnancy, were obtained from Sultan Qaboos University Hospital in Oman.

Preeclampsia is diagnosed according to the criteria of the international society for study of hypertension in pregnancy [19]. Arterial BP values $>$ 140/90 mmHg, after the 20 ${}^{\text{th}}$ week of gestation confirmed by two consecutive readings at least 6 hours apart with excess in protein-urea. Only hypertensive pregnant women between 23 to 36 weeks of gestation are included in this study. Subjects with cardiovascular disorders, arrhythmias and diabetes, and those on treatment, which affect HRV, e.g. hypnotics (drugs which depress autonomic functions) or autonomic blockers are excluded from the study. Subjects were selected to ensure that they have no history of hypertension, cardio-vascular or renal diseases before pregnancy. Only Omani pregnant women are included in this study. All HRV data are re-sampled at 1 Hz. Short recordings of 700 samples at rest are investigated.

2.2 Methods

A soft-decision wavelet based technique [13, 14, 15, 16, 17, 18] is implemented in this study. The FFT is also used for the purpose of comparison to identify normal pregnancy from preeclamptic pregnancy.

2.2.1 Soft-decision wavelet-decomposition (SD-WD) technique

To understand the soft-decision wavelet-decomposition based technique, the block-diagram of a one-stage wavelet-decomposition as shown in Fig. 1 is shown below [14].

Figure 1.

A single stage of wavelet-decomposition.

The input signal $x(n)$ of length- $N$ is first filtered by low-pass (LPF) and high-pass (HPF) filters and then down-sampled by a factor of 2 to produce both the “approximation” $a(n)$ and the “details” $d(n)$ . Assuming Hadamard-filters are used, $a(n)$ and $d(n)$ can be obtained simply by:

$\displaystyle a(n)=\frac{1}{2}[x(2n)+x(2n+1)],$ (1) $\displaystyle d(n)=\frac{1}{2}[x(2n)-x(2n+1)]$

If there is no prior knowledge about the energy distribution of the input sequence, a band-selection algorithm [14], can be used to decide (as a hard decision) which band is to be computed or kept for more processing. This method depends on the energy comparison between the low- and high-frequency subsequences after the down sampling of Fig. 1

$\displaystyle B=\sum\limits_{n=0}^{\frac{N}{2}-1}{(a(n))^{2}-(d(n))^{2}}$ (2)

According to the sign of $B$ , the decision is taken: If $B$ is positive, the low-frequency LF power is more than the high-frequency HF power and thus the LF band is considered, and if $B$ is negative, the HF band is considered. Since we are not interested in the value of $B$ , but only in its sign, a more-simpler equation than Eq. (2) can be obtained approximately as [14]:

$\displaystyle\text{sgn}(B)=\text{sgn}\sum\limits_{n=0}^{\frac{N}{2}-1}{(|a(n)|% -|d(n)|)}$ (3)

Since the hard decision method is not accurate, the soft-decision wavelet-based algorithm, which is well described in [14], can be used to estimate the power spectral density of the RRI data using the following steps:

The wavelet-decompositions shown in Fig. 1 are computed with all branches up to a certain stage $m$ to obtain 2 ${}^{m}$ sub-bands. The number of stages $m$ is selected according to the application.

All estimator results up to stage $m$ are stored, and a “probability measure” is assigned to each path (i. e., frequency band) to bear the primary information.

If $J(L)$ is the assigned probability of the input signal being primarily low-pass, the number $J(H)=1-J(L)$ is the probability that the signal is primarily high-pass. One simple way to make the probability assignments is to use the ratio of the number of positive comparisons between $|a(n)|$ and $|d(n)|$ in Eq. (3) to the total number of comparisons for a given stage.

At the following stage, the resulting estimate can be interpreted as the conditional probability of the new input sequence containing primarily low (high) frequency components, given that the previous branch was of predominantly low (high)-pass character. Using this reasoning and laws of probability, the assignments for the probability measure of the resulting sub-bands should be made equal to the product of the previous branch probability and the conditional probability estimated at a given stage. Figure 2 shows this step of probability assignment for 8 sub-bands.

The probabilities $P(B_{i})$ derived from the estimator outputs, where $i$ is the index of the band, may be interpreted themselves as a coarse measurement of the PSD [14]: The higher the probability value of any band, the higher is its power-spectral content. For $m$ decomposition stages, 2 ${}^{m}$ bands result. Each band covers 0.5/2 ${}^{m}$ Hz of the RRI-spectrum band between 0 and 0.5 Hz.

Figure 2.

PSD estimation by probability measures.

2.2.2 Implementation

In this paper, the number of stages $m$ is selected to be 5, so a total of 32 bands are obtained. The wavelet filter used is the Sym8 filter.

The estimation of the three main PSD values can be written in the following formulae according to the definition of these bands and the resolution of the algorithm (0.5/32) Hz:

$\displaystyle\overline{\textit{PSD}}_{\textit{VLF}}=\sum\limits_{i=2}^{i=3}{P(% B_{i})},$ (4) $\displaystyle\overline{\textit{PSD}}_{\textit{LF}}=\sum\limits_{i=4}^{i=10}{P(% B_{i})},$ (5) $\displaystyle\overline{\textit{PSD}}_{\textit{HF}}=\sum\limits_{i=11}^{i=25}{P% (B_{i})},$ (6)

The 40 subjects are divided equally into two groups. The first group is to be used for trial and the second group for testing. The threshold value is to be found from the trial stage and is to be used in the test stage to evaluate the efficiency of the identification.

In this study, only binary classification is considered, e.g. classification between two different cases termed “positive case” accords to preeclampsia and “negative case” accords to normal pregnancy.

The performance of a classifier is evaluated by three main metrics: Specificity, Sensitivity, and Accuracy, as follows [20]:

$\displaystyle\textit{Specificity}(\%)=\frac{\textit{TN}}{\textit{TN}+\textit{% FP}}.100$ (7) $\displaystyle\textit{Sensitivity}(\%)=\frac{\textit{TP}}{\textit{TP}+\textit{% FN}}.100$ (8) $\displaystyle\textit{Accuracy}(\%)=\frac{\textit{TP}+\textit{TN}}{\textit{TP}+% \textit{FP}+\textit{TN}+\textit{FN}}.100$ (9)

where the entities in the above equations are: (TN (true negatives), TP (true positives), FN (false negatives), and FP (false positives)).

Specificity indicates the ability of the classifier to detect negative cases. Sensitivity represents the ability of the classifier to detect the positive cases. Accuracy represents the overall performance of the classifier, which indicates the percentage of correctly classified positive and negative cases among the total number of cases.

3. Results

A complete identification system is used in this paper. The 40 subjects (20 normal and 20 preeclamptic) are divided into two equal size groups: Trail group and test group. Each group consists of 10 normal and 10 preeclamptic subjects.

Table 1 shows the mean and variance values of the three main bands of both subjects (normal and preeclamptic). It is clearly to be noted that the mean values of the VLF and LF bands are higher in preeclamptic subjects than normal subjects, while the HF power in preeclamptic subjects is less than in normal subjects.

Table 1
Mean values and variances of the three main bands of HRV of 40 subjects (20 normal and 20 preeclamptic pregnancy)

Band	Normal/mean	Normal/variance	Patient/mean	Patient/variance
VLF	0.345	0.017	0.369	0.032
LF	0.275	0.012	0.371	0.024
HF	0.354	0.077	0.317	0.104

Table 2

Specificity, sensitivity and accuracy of data groups using wavelet method

Data	Specificity	Sensitivity	Accuracy
Trial	80%	80%	80%
Test	100%	80%	90%
All	90%	80%	85%

Figure 3.

LF power results of trial data and test data using soft-decision wavelet method.

Figure 3 shows the individual values of the LF power in both preeclamptic and normal subjects. The left hand side of the figure shows the results of trial group. The threshold (thr $=$ 0.325) is found by receiver operating characteristics method (ROC) [20]. The same threshold is used on the right hand side figure to evaluate the test data. The identification results of both trial and test groups and of all data are listed in Table 2.

Figure 4 shows the individual values of the LF power in both preeclamptic and normal subjects computed using FFT spectral analysis. The left hand side of the figure shows the results of trial group. Table 3 shows the identification results using FFT for comparison with the soft-decision wavelet method.

Figure 4.

LF power results of trial data and test data using FFT method.

Table 3

Specificity, sensitivity and accuracy of data groups using FFT method

Data	Specificity	Sensitivity	Accuracy
Trial	70%	80%	75%
Test	60%	90%	75%
All	65%	85%	75%

To understand these results we need to relate them to the physiological interpretation discussed in Section 1. During pregnancy and preeclampcia, the parasympathetic activities (related to HF) are reduced and the sympathetic activities (related more to LF) are increased compared to normal. The VLF, which has till now no clear physiological interpretation, has been found to act in a similar way to the LF component.

The identification results of the soft-decision wavelet method are much better than the results of the classical FFT.

4. Conclusions

The soft-decision algorithm based on wavelet-decomposition (SD-WD) was used in this paper in a complete identification system between preeclamptic pregnancy andnormal pregnancy.

A complete system in the sense that it uses two sets of data: One for trial and the other for test.

A clear evidence of literature review regarding the behavior of the autonomic functions measured by the power of the HRV bands is obtained in this work. A reduction in the power of the HF band with a corresponding increase in the power of the LF band is obtained.

The LF band shows best result for identification. The accuracy result of 85% was obtained between preeclamptic and normal subjects with a specificity of 90% and a sensitivity of 80% on the complete set of data using the wavelet-based spectral analysis compared with an identification accuracy of 75% using FFT method.

Conflict of interest

The LF power obtained by soft-decision wavelet-decomposition is shown to be a successful feature for identification of preeclampsia.

References

Al-Hinai

. Complications of pregnancy and delivery among Omani women in different parity groups, M.Sc. Thesis at Sultan Qaboos University, College of Medicine, 2003.

Yang

Chao

Kuo

Yin

Chen

. Preeclamptic pregnancy is associated with increased sympathetic and decreased parasympathetic control of HR. Am J Physiol Heart Circ Physiol. 2000; 278: H1269-73.

Airaksinen

Kirkinen

Takkunen

. Autonomic nervous dysfunction in severe pre-eclampsia. Eur J Obstet Gynecol Reprod Biol. 1985; 19: 269-276.

Brooks

Kane

Van Winkle

. Altered heart rate baroreflex during pregnancy: Role of sympathetic and parasympatheticnervous systems. Am J Physiol. 1997; 273: R960-6.

Greenwood

Stoker

Walker

Mary

. Sympathetic nerve discharge in normal pregnancy and pregnancy-induced hypertension. J Hypertens. 1998; 16: 617-624.

Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology. Heart rate variability. Standards of measurement, physiological interpretation, and clinical use. Eur Heart J. 1996; 17: 354-381.

Gul Ar Navi Khan1, Nazia Ishrat, Zulquarnain, Analysis of heart rate variability in pre-eclamptic pregnancy: A study employing frequency domain analysis, Int J Reprod Contracept Obstet Gynecol. 2014 Dec; 3(4): 1037-1042.

Hossen

Jaju

Barhoum

Gowri

Hamdi

Hassan

Al-Kharusi

. Investigation of the High Frequency Band of Heart Rate Variability: Identification Of Preeclamptic Pregnancy from Normal Pregnancy in Oman. Asian Biomedicine Journal. 2013; 7(3): 339-346.

Hossen

Barhoum

Jaju

Gowri

Al-Kharousi

Hassan

Al-Hashmi

. Spectral Analysis of Heart Rate Variability for Normal and Preeclamptic Pregnants. International Conference on Pattern Recognition and Machine Learning, Paris-France. 30–31 December 2014; 8(12): 1971-1976.

10.

Kumari

Sharma

Kumar

. A Study of Heart Rate Variability in Pregnancy in a Tertiary Care Hospital. International Journal of Scientific Study. October 2014; 2(7).

11.

Gandhi

Mehta

Gokhale

Desai

Gokhale

Shah

. A study on cardiac autonomic modulation during pregnancy by non-invasive heart rate variability measurement. International Journal of Medicine and Public Health. 2014; 4(4): 441-445.

12.

Musa

Adam

Lutfi

. Heart Rate Variability and Autonomic Modulations in Preeclampsia. PLoS One. 2016; 11(4).

13.

Hossen

Heute

. Fully adaptive evaluation of SB-DFT, proceedings of IEEE Int. Symposium on Circuits and Systems, Chicago, Illinois, 1993.

14.

Hossen

. Power spectral density estimation via wavelet decomposition. Electronics Letters. 2004; 40: 1055-1056.

15.

Hossen

Al-Ghunaimi

Hassan

. Subband decomposition soft decision algorithm for heart rate variability analysis in patients with OSA and normal controls. Signal Processing. 2005; 85: 95-106.

16.

Hossen

. A soft decision algorithm for obstructive sleep apnea patient classification based on fast estimation of wavelet entropy of RRI data. Technol Health Care. 2005; 3: 151-165.

17.

Hossen

Al-Ghunaimi

. A wavelet-based soft decision technique for screening of patients with congestive heart failure. Biomed Signal Proces. 2007; 2: 135-143.

18.

Hossen

Muthuraman

Raethjen

Deuschl

Heute

. Discrimination of parkinsonian tremor from essential tremor by implementation of a wavelet-based soft decision technique on EMG and accelerometer signals. Biomed Signal Proces. 2005; 5: 181-188.

19.

Hermida

Ayala

Mojón

Fernández

. Time-qualified reference values for ambulatory blood pressure monitoring in pregnancy. Hypertension. 2001; 38: 746-752.

20.

Rangayyan

. Biomedical signal analysis: A case-study approach. IEEE. NJ; 2001.

Identification of patients with preeclampsia from normal subjects using wavelet-based spectral analysis of heart rate variability

Abstract

BACKGROUND:

OBJECTIVE:

METHODS:

RESULTS:

CONCLUSION:

Keywords

1. Introduction

2. Data acquisition

2.1 Subjects

2.2 Methods

2.2.1 Soft-decision wavelet-decomposition (SD-WD) technique

Table 1 Mean values and variances of the three main bands of HRV of 40 subjects (20 normal and 20 preeclamptic pregnancy)

Conflict of interest

References

Table 1
Mean values and variances of the three main bands of HRV of 40 subjects (20 normal and 20 preeclamptic pregnancy)