Short time cardio-vascular pulses estimation for dengue fever screening via continuous-wave Doppler radar using empirical mode decomposition and continuous wavelet transform

Abstract

Contactless measurement of cardio-vascular pulse acts an essential role in clinical medical sectors. Estimating cardio-vascular pulses based on continuous-wave (CW) Doppler radar in limited time while maintaining the accuracy is a challenging task. In this paper, we propose a signal processing method that combines empirical mode decomposition (EMD) and continuous wavelet transform (CWT) for a short time estimation of heart rate (HR) and inter-beat-interval from radar signals. We evaluate performance of the proposed method using 85 patients with dengue fever and 40 healthy subjects. Subsequently, the estimated contactless HR is compared to that of a commercial contact-type medical device. The result shows that the HR can be estimated within a period of 5 s with an accuracy of 96.2 ± 2.5%. The patients with dengue fever show an elevated HR and a decreased standard deviation of heartbeat interval (SDHI). Finally, linear discriminant analysis (LDA) is utilized with two parameters HR and SDHI to classify the dengue patients, achieving the sensitivity and specificity values of 86.3% and 86.9%, respectively.

Introduction

Dengue is an infectious disease, commonly occurs in tropical and subtropical countries, which has been a global threat in the first two decades of the 21st century [1]. It is estimated that there are 390 million cases of the annual occurrence [2]. Hence, dengue screening is important to minimize the possibility of dengue fever outbreaks. Body temperature measurement devices are widely used in public areas for the screening; however, several recent studies have indicated that changing body temperature is not only caused by fever symptoms but also by antipyretic agents and ambient temperature, and thus it is insufficiently reliable [3], [4]. Beside body temperature, heart rate is another important information for the infection screening. There have been several studies reported of rapid HRs while having the symptoms of dengue fever and of decreasing HRs during recovery [5], [6]. Moreover, some recent studies have suggested that there is a difference in standard deviation of heartbeat interval (SDHI) of dengue patient and healthy person [7], [8].

Although contact method for HR measurement such as electrocardiography (ECG) and photoplethysmography (PPG) have shown highly reliable results [9], they may contain several drawbacks in clinical practice. For instance, the contact methods may be limited due to privacy and inconvenient use, making them difficult to be applied at a large scale. In addition, the clinical staffs have to reach the patients for the device attachment, potentially causing them being exposed to a high risk, especially in case of highly contagious infections. Therefore, research for contactless measurement methods for HR detection has been highly active, such as in public healthcare. In screening infectious diseases such as dengue fever, a contactless system, which is based on Doppler radar, can be used to detect HR and to classify infected patients [10]. The principle of Doppler radar-based screening is that the wave is capable to penetrate through objects such as clothes, and then rebounds with the information of the displacement of wall chest induced by the heart activity and the respiratory motion [11], [12]. The contactless system, based on CW Doppler radar, has several advantages such as low power consumption, large detection range and simple radio architecture [11], [13]. Several researches on HR estimation using Doppler radar have been published recently. Ohtsuki et al. [14] proposed a method based on estimating the average R-R interval applying the Viterbi algorithm for heartbeat detection; yet, the study assumed that the R-R intervals are almost equal within a short period, which is not appropriate for the cases with varying R-R interval. Yu et al. [15] introduced a method using higher order cyclostationary (HOCS) and proved that the third-order cyclic cumulant can detect the HR. Le [16] proposed a method to improve the SNR of the heartbeat signal based on high order cumulant (HOC). Subsequently, Z-chirp transform was applied to zoom in a specific band of frequency to extract the HR. Saluja et al. [17] applied a gamma filter, a supervised machine learning algorithm, to eliminate the respiratory signal and then extract the heartbeat signal; nevertheless, it is just feasible to extract the HR from the radar signals of arbitrary subjects with reasonable accuracy when a vast amount of data are involved in training the model.

HR estimation in a short time is crucial in several clinical applications, including estimating the vital signs of patient at the registration desk in health care facilities (especially worthy in emergency cases), and fast screening passengers from epidemic areas which may indicate the probability of infection, requiring a short determination time (typically within 5 s or less) [18]. Some conventional methods such as Fourier transform (FFT), HOCS and HOC are not suitable for fast estimation of cardio-vascular pulse, because these methods require a period of signal longer than 10 s to convert to the frequency domain and highlight the frequency composition of the heartbeat. Some recent studies have been focused on cardio-vascular pulse estimation in a short time using Doppler radar signal. Yang et al. [19] utilized a peak-detection algorithm in the time domain to measure the HR extracted from the radar signal by applying two band pass filters (BPF) in approximately 5 s, and obtained the correlation coefficient of 0.92 between the raw extracted heartbeat signal and a reference ECG. Furthermore, several methods based on data-length-variation techniques have been proposed to estimate the HR in a short time. Tu and Lin [20] studied 4 healthy subjects using FFT and showed that the HR can be determined within time variation of from 2 to 5 s with an error of 3.4%. Li and Lin [18] applied a method with the wavelet transform coefficients on data from 3 subjects, and achieved an error of less than 4% within window times from 3 to 5 s. Those mentioned methods, however, analyze the HR in the frequency domain and thus are not suitable to extract SDHI information. In addition, the mentioned studies just carried on a limited amount of healthy subjects, and thus their results are still unfirmly solid with dengue patients.

Typically, the displacement of the wall chest caused by the breathing is naturally often stronger than the displacement of the heart activity [18], [21]. Also, CW Doppler radar is widely confirmed that the received signal has non-stationary and non-linear characteristics [22]. As the result, separating the heartbeat signal from the output of continuous-wave Doppler radar is challenging and garners the attention of researchers. Based on the fact that heartbeat signal can be found in the specific frequency range from 0.5 to 2.5 Hz [23], there have been several studies related to heartbeat signal separation using BPF [19], [24], wavelet transform (WT) [18], [25] and empirical mode decomposition (EMD) [26]. Droitcour et al. [24] used a two-stage BPF, in which, the first stage BPF is to reduce high-frequency noise while the second stage is to extract the heartbeat signal. While BPF is not suitable method for processing the non-linear and non-stationary signal, WT and EMD may be used to handle this problem. He and Liu [25] proposed a method utilizing WT with the frequency range from 0.8 to 1.5 Hz, and the heartbeat signal retrieved by wavelet reconstruction. Mostafanezhad et al. [26] applied an EMD-based method to eliminate high frequency motion noise. Specifically, the received signal has a large respiratory amplitude and motion noise, thus a high-pass filter was used to attenuate the respiratory frequency before utilizing the EMD method for the heartbeat signal extraction. Subsequently, a window-based FFT method was applied to estimate the HR. Besides, Hu and Jin [27] used an improved ensemble empirical mode decomposition (EEMD) to remove the high frequency noise in the signal and then applied CWT to extract the heartbeat presented by a single frequency scale. Principle component analysis (PCA) has been utilized to extract cardio-vascular pulse patterns from the received signal [28]. PCA, however, is not suitable for signals that contain non-Gaussian noises, especially the burst noise. To the best of our knowledge, considering the paucity of studies on short time cardio-vascular pulses estimation on dengue fever infection screening, it is necessary to investigate insightfully on this problem.

In this study, we aim to classify dengue patients in a short period using the HR and SDHI information extracted by processing signal from CW Doppler radar. Constitutively, we propose a coarse-to-fine approach that combines the EMD and CWT to estimate the cardio-vascular pulse via CW Doppler radar during a short period. EMD is used for decomposing natural signals into intrinsic mode functions (IMFs). Hence, it is able to decompose and estimate components of short sample length accurately in the time domain, which can separate the heartbeat from the signal [26], [29]. In addition, due to the fact that the heartbeat signal can be distorted, CWT is used to provide a time-frequency map of the signal being analyzed with high frequency resolution and sharp time resolutions [30]. In this work, EMD is firstly used to decompose the I/Q signals into IMFs. Next, the most resemble IMF to the reference heat beat signal is selected based on the maximum of the normalized cross-correlation between each IMF and the reference signal in a pilot study. Afterwards, in order to eliminate the incorrect heart peaks in the chosen IMF, a scale-limited CWT reconstruction is applied to reform a new signal (reconstructed IMF). Subsequently, based on peak detection algorithm [31] on the reconstructed signal, the HR and SDHI information can be extracted in the time domain. Finally, we apply linear discriminant analysis (LDA) method to classify dengue patients versus healthy subjects using 85 clinical data from hospital and 40 healthy voluntaries.

The rest of this paper is organized as follows. In Section 2 the material and method section, we introduce the contactless system used for data acquisition and the proposed signal processing method to estimate cardio-vascular pulses and the HR from the received signals. Section 3 presents evaluation results of the proposed method, compared to HR extraction by a conventional method and from a commercial contact device. Also, classification of the dengue patients using LDA with the HR and SDHI is mentioned in this section. These achieved results are discussed in Section 4 while conclusions are drawn in Section 5.