Non-invasive diagnosis of fetal arrhythmia based on multi-domain feature and hierarchical extreme learning machine

https://doi.org/10.1016/j.bspc.2022.104191Get rights and content

Highlights

  • A novel deep learning framework based on H-ELM is proposed for fetal arrhythmia diagnosis.

  • A novel multi-domain feature extraction technique is proposed to characterize the fetal arrhythmia.

  • Neighborhood component analysis is adopted to screen sensitive features.

Abstract

Heart disease is one of the major causes of affecting the health of newborns. Detecting the presence or potential heart disease of the fetus as soon as possible, and adopting relevant treatment plans in a timely manner, which has a profound impact on doctors and patients. This study aims to develop an accurate screening method with arrhythmia (ARR) to assist physicians to further diagnose whether them have heart disease. Therefore, this paper proposes a multi-domain feature extraction technique and a hierarchical extreme learning machine (H-ELM) network for the prediction of fetal ARR. Firstly, the multi-domain feature extraction technology is used to extract abundant high-dimensional feature for representing the original signal. Secondly, neighborhood component analysis (NCA) is used to screen the sensitive features from the high dimensional feature vectors. Then, the obtained sensitive features are input into stacked extreme learning machine sparse autoencoder (ELM-SAE), which extract high-level fusion features by layer-by-layer unsupervised learning manner. Finally, an original ELM was connected on the end of the ELM-SAE network for the prediction of fetal ARR. The experimental results illustrate that the proposed method can achieve sensitivity of 99.11%, specificity of 93.91%, precision of 93.52%, and accuracy of 96.33%. Furthermore, the proposed method comprehensive performance outperforms the compared models. Therefore, the proposed method can be effectively used for the prediction of fetal ARR, and with the continuous improved the research, it is expected to be considered as an auxiliary diagnostic tool for physicians in the future.

Introduction

Fetal ARR are defined as tachycardia (sustained heart rate > 200 bpm), bradycardia (sustained heart rate < 100 bpm), or irregular heart rhythm [1]. The probability of fetal arrhythmia is approximately 1 % [2]. Causes of the above disorders include ischemia, inflammation, electrolyte disturbances, structural defects, and genetic conditions. While most fetal ARR are benign, about 10 % are thought to predispose to underlying diseases such as fetal hydrops and fetal death [3], [4]. Therefore, monitoring and diagnosing fetal heart rhythm, and developing appropriate ARR treatment plan in a timely manner can improve the health status of newborns.

With the advancement of medicine, fetal heart examination is no longer limited to traditional stethoscope listening to heart sounds. Fetal magnetocardiography is a non-invasive technique for recording the magnetic field generated by the electrical activity of the fetal heart. The advantage of this method is that the parent noise is small and the signal quality is good. The disadvantage is that it needs to use a magnetic shielding room to improve the signal-to-noise ratio, which is expensive [5]. Fetal echocardiography is a technique for imaging the fetal heart based on the theory of Doppler ultrasound, which can visually detect the structural condition of the fetal heart. Continuous echocardiographic recordings are usually short, limitations of this method also include the inability to detect heart rate trend during tachycardia or bradycardia [6], [7]. Non-invasive fetal electrocardiography (NI-FECG) is a promising method for non-invasive fetal heart monitoring, using multiple transabdominal electrodes to obtain fetal cardiac potentials from the maternal abdominal surface [8], [9]. The findings demonstrate that NI-FECG can accurately estimate fetal heart rate. NI-FECG has significant advantages such as low cost, non-invasiveness, remote monitoring, and acquisition of fetal heart rate trends, and this method is likely to be widely used in clinical practice [10], [11].

Due to the rich information of NI-FECG, many scholars have devoted themselves to researching different filtering methods and feature extraction methods to eliminate maternal noise from the original signal, and then extract the real signal of fetal ECG for the diagnosis of fetal heart rhythm conditions. The most commonly used methods including independent component analysis (ICA), which decomposes the original signal into statistically independent source components [12], [13]. Principal component analysis can transform the original mixed signal into several new signals that are independent of each other [14], [15]. However, the fECG component of the original signal is weak and difficult to find among the various components. Wavelet Transform (WT) [16] is used to analyze the time–frequency domain characteristics of the original signal, but it cannot fully reflect the waveform characteristics of the original signal. Empirical mode decomposition (EMD) is used to disassemble the signal into a set of intrinsic mode functions (IMFs) [17], [18], which cannot overcome mode aliasing. Recent studies have shown that multi-domain feature outperforms the single-domain feature in the classification task. Barnova et al. [19] combined ICA, recursive least squares, and ensemble empirical mode decomposition algorithms to extract the QRS waves of fetal ECG signals from abdominal recordings. Wu et al. [20] proposed a fetal ECG extraction method based on WT, least mean square adaptive filtering algorithm and spatial selective noise filtering algorithm, which can reconstruct fetal electrocardiogram (fECG) after denoising. Suganthy et al. [21] performed Savitzky-Golay filtering and symlet wavelet transform on the original abdominal signal to eliminate the baseline offset noise, and used an adaptive RLS filter to extract the fECG with reference to the mother's chest ECG. Panigrahy et al. [22] used the adaptive neuro-fuzzy inference system framework, and added a differential evolution algorithm to the Extended Kalman Smoothing algorithm to extract fECG signals from abdominal ECG signals. The above research can effectively extract the fECG signal fragment from the original signal, but it still needs professional medical personnel to diagnose one by one.

Recently, some traditional machine learning algorithms, such as extreme learning machine (ELM) [23], support vector machine (SVM) [24] and so on, have been successfully applied to ARR diagnosis, and they are expected to be an effective tool to assist medical staff in the diagnosis of patients. However, there are still the following issues in the diagnosis of fetal ARR: 1) Blind source separation technology is widely used to separate fECG signals, but it unable to provide a qualitative result. It still needs medical staff to make diagnosis one by one. 2) Only using single-domain features as the description of the original signal cannot fully capture the characteristic information of the original signal. 3) Although shallow neural networks have achieved certain results in fetal arrhythmia diagnosis, most these algorithms do not have the ability to self-learn high-level and meaningful features from input data.

To tackle above problems, the multi-domain features extraction technology and the H-ELM [25] network is proposed for diagnosing fetal heart rhythm, which aims to use machine learning algorithm to mine the implicit connection of data and realize the intelligent classification of ARR. H-ELM is a novel hierarchical learning framework of multi-layer perceptron, which adopted ELM as the classifier and used ELM-based sparse auto-encoder, unsupervised multi-layer encoding for feature extraction between the input layer and the classifier, to obtain high-level features that are more compact than the original features layer by layer. The flowchart of the proposed method is shown in Fig. 1. First, we developed the multi-domain feature extraction technique that comprehensively exploits the advantages of the statistical analysis, wavelet entropy, sample entropy and Hilbert-Huang transform (HHT) algorithm to extract rich feature information from the original signal. Second, NCA is used to screen the sensitive features from the high dimensional feature vectors. Then, the obtained sensitive features are input into stacked ELM-SAE. Finally, an original ELM was connected on the end of the ELM-SAE network for the prediction of fetal ARR.

The remainder of this paper is as follows. Multi-domain feature extraction techniques and H-ELM network structures are described in detail in Section 2. Experimental method and result are depicted in Section 3. The advantage and defect of this paper are discussed in Section 4. Section 5 is the conclusion of this paper and conception of future work.

Section snippets

Multi-domain feature

The existing fetal arrhythmia diagnosis technology generally focuses on the classifier model with single-domain features, which easily leads to insufficient feature extraction and low identification accuracy [26], [27]. Therefore, we developed the multi-domain feature extraction technique that comprehensively exploits the advantages of the statistical analysis, wavelet entropy, sample entropy and HHT algorithms to extract rich feature information from the original signal.

Statistical analysis,

Data description

The Non-invasive fetal arrhythmia database was developed by the Technion-Israel Institute of Technology. It is freely available on the PhysioNet website (https://physionet.org/physiobank/database/nifeadb/). NIFEADB included 12 cases diagnosed with ARR and 14 cases with fetal normal heart rhythm (NR). NI‐FECG uses Cardiolab Babycard equipment to acquire data recordings through 5 or 6 electrodes placed on the abdomen of the pregnant woman (one of which is a common reference) and 2 chest

Discussion

In this study, a deep network model was established to achieve the classification and prediction of ARR, which provided a basis for further diagnosis. This study has several limitations. First, this paper only achieved the classification of fetal rhythm status, without carry out an in-depth study on the possible disease types of ARR. Secondly, the contribution degree of each channel signal to classification results is not studied in this paper, resulting in large redundancy of original

Conclusion

To address insufficient feature extraction and low identification accuracy, we propose a novel multi-domain feature extraction technique and H-ELM for the prediction of fetal ARR. Firstly, the multi-domain feature extraction technology is used to extract abundant high-dimensional features from the raw signal. Secondly, NCA is used to screen the sensitive feature from the high dimensional feature sets. Then, the obtained sensitive feature vectors are input into stacked ELM-SAE, which obtains

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments

This research was funded by Sichuan Science and Technology Program, grant number 2021YFS0065.

References (40)

  • J. Jezewski, K. Horoba, A. Matonia, A. Gacek, M. Bernys, A new approach to cardiotocographic fetal monitoring based on...
  • J.A. Behar et al.

    Noninvasive fetal electrocardiography for the detection of fetal arrhythmias

    Prenat. Diagn.

    (2019)
  • G.D. Clifford et al.

    Non-invasive fetal ECG analysis

    Physiol. Meas.

    (2014)
  • L. Yuan, Z. Zhou, Y. Yuan, S. Wu, An Improved FastICA Method for Fetal ECG Extraction, Comput. Math. Methods Med. 2018...
  • J. Giraldo-Guzmán et al.

    Fetal ECG extraction using independent component analysis by Jade approach

  • R. Martinek et al.

    Comparative Effectiveness of ICA and PCA in Extraction of Fetal ECG From Abdominal Signals: Toward Non-invasive Fetal Monitoring

    Front. Physiol.

    (2018)
  • R. Petrolis et al.

    Multistage principal component analysis based method for abdominal ECG decomposition

    Physiol. Meas.

    (2015)
  • K.D. Desai et al.

    A real-time fetal ECG feature extraction using multiscale discrete wavelet transform

  • P. Ghobadi Azbari et al.

    A novel approach to the extraction of fetal electrocardiogram based on empirical mode decomposition and correlation analysis

    Australas. Phys. Eng. Sci. Med.

    (2017)
  • G. Liu et al.

    An adaptive integrated algorithm for noninvasive fetal ECG separation and noise reduction based on ICA-EEMD-WS

    Med. Biol. Eng. Comput.

    (2015)
  • Cited by (6)

    • Optimizing the Capacity of Extreme Learning Machines for Biomedical Informatics Applications

      2023, 1st International Conference on Emerging Research in Computational Science, ICERCS 2023 - Proceedings
    1

    The first two authors contributed equally to this work and are considered co-first authors.

    View full text