New Marginal Spectrum Feature Information Views of Humpback Whale Vocalization Signals Using the EMD Analysis Methods.

Marginal spectrum (MS) feature information of humpback whale vocalization (HWV) signals is an interesting and significant research topic. Empirical mode decomposition (EMD) is a powerful time-frequency analysis tool for marine mammal vocalizations. In this paper, new MS feature innovation information of HWV signals was extracted using the EMD analysis method. Thirty-six HWV samples with a time duration of 17.2 ms were classified into Classes I, II, and III, which consisted of 15, 5, and 16 samples, respectively. The following ratios were evaluated: the average energy ratios of the 1 first intrinsic mode function (IMF1) and residual function (RF) to the referred total energy for the Class I samples; the average energy ratios of the IMF1, 2nd IMF (IMF2), and RF to the referred total energy for the Class II samples; the average energy ratios of the IMF1, 6th IMF (IMF6), and RF to the referred total energy for the Class III samples. These average energy ratios were all more than 10%. The average energy ratios of IMF1 to the referred total energy were 9.825%, 13.790%, 4.938%, 3.977%, and 3.32% in the 2980-3725, 3725-4470, 4470-5215, 10,430-11,175, and 11,175-11,920 Hz bands, respectively, in the Class I samples; 14.675% and 4.910% in the 745-1490 and 1490-2235 Hz bands, respectively, in the Class II samples; 12.0640%, 6.8850%, and 4.1040% in the 2980-3725, 3725-4470, and 11,175-11,920 Hz bands, respectively, in the Class III samples. The results of this study provide a better understanding, high resolution, and new innovative views on the information obtained from the MS features of the HWV signals.

Empirical Mode Decomposition-Based Feature Extraction for Environmental Sound Classification.

In environment sound classification, log Mel band energies (MBEs) are considered as the most successful and commonly used features for classification. The underlying algorithm, fast Fourier transform (FFT), is valid under certain restrictions. In this study, we address these limitations of Fourier transform and propose a new method to extract log Mel band energies using amplitude modulation and frequency modulation. We present a comparative study between traditionally used log Mel band energy features extracted by Fourier transform and log Mel band energy features extracted by our new approach. This approach is based on extracting log Mel band energies from estimation of instantaneous frequency (IF) and instantaneous amplitude (IA), which are used to construct a spectrogram. The estimation of IA and IF is made by associating empirical mode decomposition (EMD) with the Teager-Kaiser energy operator (TKEO) and the discrete energy separation algorithm. Later, Mel filter bank is applied to the estimated spectrogram to generate EMD-TKEO-based MBEs, or simply, EMD-MBEs. In addition, we employ the EMD method to remove signal trends from the original signal and generate another type of MBE, called S-MBEs, using FFT and a Mel filter bank. Four different datasets were utilised and convolutional neural networks (CNN) were trained using features extracted from Fourier transform-based MBEs (FFT-MBEs), EMD-MBEs, and S-MBEs. In addition, CNNs were trained with an aggregation of all three feature extraction techniques and a combination of FFT-MBEs and EMD-MBEs. Individually, FFT-MBEs achieved higher accuracy compared to EMD-MBEs and S-MBEs. In general, the system trained with the combination of all three features performed slightly better compared to the system trained with the three features separately.

A novel heart sound segmentation algorithm via multi-feature input and neural network with attention mechanism.

Objective. Heart sound segmentation (HSS), which aims to identify the exact positions of the first heart sound(S1), second heart sound(S2), the duration of S1, systole, S2, and diastole within a cardiac cycle of phonocardiogram (PCG), is an indispensable step to find out heart health. Recently, some neural network-based methods for heart sound segmentation have shown good performance.Approach. In this paper, a novel method was proposed for HSS exactly using One-Dimensional Convolution and Bidirectional Long-Short Term Memory neural network with Attention mechanism (C-LSTM-A) by incorporating the 0.5-order smooth Shannon entropy envelope and its instantaneous phase waveform (IPW), and third intrinsic mode function (IMF-3) of PCG signal to reduce the difficulty of neural network learning features.Main results. An average F1-score of 96.85 was achieved in the clinical research dataset (Fuwai Yunnan Cardiovascular Hospital heart sound dataset) and an average F1-score of 95.68 was achieved in 2016 PhysioNet/CinC Challenge dataset using the novel method.Significance. The experimental results show that this method has advantages for normal PCG signals and common pathological PCG signals, and the segmented fundamental heart sound(S1, S2), systole, and diastole signal components are beneficial to the study of subsequent heart sound classification.

Analysis of Pressure Fluctuation Characteristics of Central Swirl Combustors Based on Empirical Mode Decomposition.

In order to study the characteristics of pressure fluctuation during unstable combustion, experimental studies had been conducted on the mechanism model of the swirl combustor and the industrial swirl combustor. The signal of dynamic pressure, heat release rate, and the high-speed flame image in the two combustors were synchronously collected by using dynamic pressure sensors, a photoelectric sensor, and a high-speed camera under normal temperature and pressure. After empirical mode decomposition of the dynamic pressure signal, several intrinsic mode functions were obtained. It was found that the pressure pulsation energy is concentrated in the first three order intrinsic mode function. Through fast Fourier transform spectrum calculation, it was found that the first three order intrinsic mode function pulsation can characterize the changes of heat release rate and air flow pulsation under cold state and flame instability. It showed that the decomposition of the dynamic pressure in the combustor by this method can obtain the main physical processes in its connotation, and provide data processing methods for the induction mechanism of oscillating combustion and combustion diagnosis in an industrial combustor test.

Hybrid-Pattern Recognition Modeling with Arrhythmia Signal Processing for Ubiquitous Health Management.

We established a web-based ubiquitous health management (UHM) system, "ECG4UHM", for processing ECG signals with AI-enabled models to recognize hybrid arrhythmia patterns, including atrial premature atrial complex (APC), atrial fibrillation (AFib), ventricular premature complex (VPC), and ventricular tachycardia (VT), versus normal sinus rhythm (NSR). The analytical model coupled machine learning methods, such as multiple layer perceptron (MLP), random forest (RF), support vector machine (SVM), and naive Bayes (NB), to process the hybrid patterns of four arrhythmia symptoms for AI computation. The data pre-processing used Hilbert-Huang transform (HHT) with empirical mode decomposition to calculate ECGs' intrinsic mode functions (IMFs). The area centroids of the IMFs' marginal Hilbert spectrum were suggested as the HHT-based features. We engaged the MATLABTM compiler and runtime server in the ECG4UHM to build the recognition modules for driving AI computation to identify the arrhythmia symptoms. The modeling extracted the crucial data sets from the MIT-BIH arrhythmia open database. The validated models, including the premature pattern (i.e., APC-VPC) and the fibril-rapid pattern (i.e., AFib-VT) against NSR, could reach the best area under the curve (AUC) of the receiver operating characteristic (ROC) of approximately 0.99. The models for all hybrid patterns, without VPC versus AFib and VT, achieved an average accuracy of approximately 90%. With the prediction test, the respective AUCs of the NSR and APC versus the AFib, VPC, and VT were 0.94 and 0.93 for the RF and SVM on average. The average accuracy and the AUC of the MLP, RF, and SVM models for APC-VT reached the value of 0.98. The self-developed system with AI computation modeling can be the backend of the intelligent social-health system that can recognize hybrid arrhythmia patterns in the UHM and home-isolated cares.

Deep-learning online EEG decoding brain-computer interface using error-related potentials recorded with a consumer-grade headset.

Objective.Brain-computer interfaces (BCIs) allow subjects with sensorimotor disability to interact with the environment. Non-invasive BCIs relying on EEG signals such as event-related potentials (ERPs) have been established as a reliable compromise between spatio-temporal resolution and patient impact, but limitations due to portability and versatility preclude their broad application. Here we describe a deep-learning augmented error-related potential (ErrP) discriminating BCI using a consumer-grade portable headset EEG, the Emotiv EPOC+.Approach.We recorded and discriminated ErrPs offline and online from 14 subjects during a visual feedback task.Main results:We achieved online discrimination accuracies of up to 81%, comparable to those obtained with professional 32/64-channel EEG devices via deep-learning using either a generative-adversarial network or an intrinsic-mode function augmentation of the training data and minimalistic computing resources.Significance.Our BCI model has the potential of expanding the spectrum of BCIs to more portable, artificial intelligence-enhanced, efficient interfaces accelerating the routine deployment of these devices outside the controlled environment of a scientific laboratory.

Energy-Period Profiles of Brain Networks in Group fMRI Resting-State Data: A Comparison of Empirical Mode Decomposition With the Short-Time Fourier Transform and the Discrete Wavelet Transform.

Traditionally, functional networks in resting-state data were investigated with linear Fourier and wavelet-related methods to characterize their frequency content by relying on pre-specified frequency bands. In this study, Empirical Mode Decomposition (EMD), an adaptive time-frequency method, is used to investigate the naturally occurring frequency bands of resting-state data obtained by Group Independent Component Analysis. Specifically, energy-period profiles of Intrinsic Mode Functions (IMFs) obtained by EMD are created and compared for different resting-state networks. These profiles have a characteristic distribution for many resting-state networks and are related to the frequency content of each network. A comparison with the linear Short-Time Fourier Transform (STFT) and the Maximal Overlap Discrete Wavelet Transform (MODWT) shows that EMD provides a more frequency-adaptive representation of different types of resting-state networks. Clustering of resting-state networks based on the energy-period profiles leads to clusters of resting-state networks that have a monotone relationship with frequency and energy. This relationship is strongest with EMD, intermediate with MODWT, and weakest with STFT. The identification of these relationships suggests that EMD has significant advantages in characterizing brain networks compared to STFT and MODWT. In a clinical application to early Parkinson's disease (PD) vs. normal controls (NC), energy and period content were studied for several common resting-state networks. Compared to STFT and MODWT, EMD showed the largest differences in energy and period between PD and NC subjects. Using a support vector machine, EMD achieved the highest prediction accuracy in classifying NC and PD subjects among STFT, MODWT, and EMD.

Empirical Mode Decomposition-Based Method for Artefact Removal in Raw Intracranial Pressure Signals.

Intracranial pressure (ICP) signals are often contaminated by artefacts and segments of missing values. Some of these artefacts can be observed as very high and short spikes with a physiologically impossible high slope. The presence of these spikes reduces the accuracy of pattern recognition techniques. Thus, we propose a modified empirical mode decomposition (EMD) method for spike removal in raw ICP signals. The EMD breaks down the signal into 16 intrinsic mode functions (IMFs), combines the first 4 to localize spikes using adaptive thresholding, and then either removes or imputes the identified ICP spikes.

Estimating Heart Rate and Respiratory Rate from a Single Lead Electrocardiogram Using Ensemble Empirical Mode Decomposition and Spectral Data Fusion.

Cardiopulmonary monitoring is important and useful for diagnosing and managing multiple conditions, such as stress and sleep disorders. Wearable ambulatory systems can provide continuous, comfortable, and inexpensive means for monitoring; it always has been a research subject in recent years. Being simple and cost-effective, electrocardiogram-based commercial products can be found in the market that provides cardiac diagnostic information for assessment, including heart rate measurement and atrial fibrillation identification. Based on a data-driven and self-adaptive approach, this study aims to estimate heart rate and respiratory rate simultaneously from one lead electrocardiogram signal. In contrast to ensemble empirical mode decomposition with principle component analysis, performed in the time domain, our method uses spectral data fusion, together with intrinsic mode functions using ensemble empirical mode decomposition obtains a more accurate heart rate and respiratory rate. Equipped with a rule-based selection of defined frequency levels for respiratory rate (RR) estimation, the proposed method obtains (0.92, 1.32) beat per minute for the heart rate and (2.20, 2.92) breath per minute for the respiratory rate as their mean absolute error and root mean square error, respectively outperforming other existing methods.

Using an Optimization Algorithm to Detect Hidden Waveforms of Signals.

Source signals often contain various hidden waveforms, which further provide precious information. Therefore, detecting and capturing these waveforms is very important. For signal decomposition (SD), discrete Fourier transform (DFT) and empirical mode decomposition (EMD) are two main tools. They both can easily decompose any source signal into different components. DFT is based on Cosine functions; EMD is based on a collection of intrinsic mode functions (IMFs). With the help of Cosine functions and IMFs respectively, DFT and EMD can extract additional information from sensed signals. However, due to a considerably finite frequency resolution, EMD easily causes frequency mixing. Although DFT has a larger frequency resolution than EMD, its resolution is also finite. To effectively detect and capture hidden waveforms, we use an optimization algorithm, differential evolution (DE), to decompose. The technique is called SD by DE (SDDE). In contrast, SDDE has an infinite frequency resolution, and hence it has the opportunity to exactly decompose. Our proposed SDDE approach is the first tool of directly applying an optimization algorithm to signal decomposition in which the main components of source signals can be determined. For source signals from four combinations of three periodic waves, our experimental results in the absence of noise show that the proposed SDDE approach can exactly or almost exactly determine their corresponding separate components. Even in the presence of white noise, our proposed SDDE approach is still able to determine the main components. However, DFT usually generates spurious main components; EMD cannot decompose well and is easily affected by white noise. According to the superior experimental performance, our proposed SDDE approach can be widely used in the future to explore various signals for more valuable information.

Fault Diagnosis for Rolling Bearings Using Optimized Variational Mode Decomposition and Resonance Demodulation.

It is difficult to extract the fault signal features of locomotive rolling bearings and the accuracy of fault diagnosis is low. In this paper, a novel fault diagnosis method based on the optimized variational mode decomposition (VMD) and resonance demodulation technology, namely GNVRFD, is proposed to realize the fault diagnosis of locomotive rolling bearings. In the proposed GNVRFD method, the genetic algorithm and nonlinear programming are combined to design a novel parameter optimization algorithm to adaptively optimize the two parameters of the VMD. Then the optimized VMD is employed to decompose the collected vibration signal into a series of intrinsic mode functions (IMFs), and the kurtosis value of each IMF is calculated, respectively. According to the principle of maximum value, two most sensitive IMF components are selected to reconstruct the vibration signal. The resonance demodulation technology is used to decompose the reconstructed vibration signal in order to obtain the envelope spectrum, and the fault frequency of locomotive rolling bearings is effectively obtained. Finally, the actual data of rolling bearings is selected to testify the effectiveness of the proposed GNVRFD method. The experiment results demonstrate that the proposed GNVRFD method can more accurately and effectively diagnose the fault of locomotive rolling bearings by comparing with other fault diagnosis methods.

Revised Stability Scales of the Postural Stability Index for Human Daily Activities.

Evaluation of human postural stability is important to prevent falls. Recent studies have been carried out to develop postural stability evaluation in an attempt to fall prevention. The postural stability index (PSI) was proposed as a measure to evaluate the stability of human postures in performing daily activities. The objective of this study was to use the PSI in developing the stability scales for human daily activities. The current study used two open datasets collected from mobile devices. In addition, we also conducted three experiments to evaluate the effect of age, velocity, step counts, and devices on PSI values. The collected datasets were preprocessed using the ensemble empirical mode decomposition (EEMD), then the complexity index from each intrinsic mode function (IMF) was calculated using the multiscale entropy (MSE). From the evaluation, it can be concluded that the PSI can be applied to do daily monitoring of postural stability for both young and older adults, and the PSI is not affected by age. The revised stability scales developed in this current study can give better suggestions to users than the original one.

Epileptic seizure classifications using empirical mode decomposition and its derivative.

BACKGROUND: Epilepsy is one of the most common neurological disorders associated with disruption of brain activity. In the classification and detection of epileptic seizures, electroencephalography (EEG) measurements, which record the electrical activities of the brain, are frequently used. Empirical mode decomposition (EMD) and its derivative, ensemble EMD (EEMD) are recently developed methods used to decompose non-stationary and nonlinear signals such as EEG into a finite number of oscillations called intrinsic mode functions (IMFs). Our main objective in this study is to present a hybrid IMF selection method combining four different approaches (energy, correlation, power spectral distance, and statistical significance measures), and investigate the effect of selected IMFs extracted by EMD and EEMD on the classification. We have applied the proposed IMF selection approach on the classification of EEG signals recorded from epilepsy patients who are under treatment at our collaborator hospital. Multichannel EEG signals collected from epilepsy patients are decomposed into IMFs, and then IMF selection was performed. Finally, time- and spectral-domain, and nonlinear features are extracted and feature sets are created for the classification. RESULTS: The maximum classification accuracies obtained using various combinations of IMFs were 94.56%, 95.63%, 96.8%, and 96.25% for SVM, KNN, naive Bayes, and logistic regression classifiers, respectively, by using EMD analysis; whereas, the EEMD approach has provided maximum classification accuracies of 96.06%, 97%, 97%, and 96.25% for SVM, KNN, naive Bayes, and logistic regression, respectively. Classification performance with the same features obtained using direct EEG signals instead of the decomposed IMFs was worse than the aforementioned 2 approaches for every combination. CONCLUSION: Simulation results demonstrate that the proposed IMF selection approach affects the classification results. Also, EEMD provides a robust method for feature extraction from EEG signals in order to classify pre-seizure and seizure segments.

Intrinsic Frequencies of the Resting-State fMRI Signal: The Frequency Dependence of Functional Connectivity and the Effect of Mode Mixing.

The frequency characteristics of the resting-state BOLD fMRI (rs-fMRI) signal are of increasing scientific interest, as we discover more frequency-specific biological interpretations. In this work, we use variational mode decomposition (VMD) to precisely decompose the rs-fMRI time series into its intrinsic mode functions (IMFs) in a data-driven manner. The accuracy of the VMD decomposition of constituent IMFs is verified through simulations, with higher reconstruction accuracy and much-reduced mode mixing relative to previous methods. Furthermore, we examine the relative contribution of the VMD-derived modes (frequencies) to the rs-fMRI signal as well as functional connectivity measurements. Our primary findings are: (1) The rs-fMRI signal within the 0.01-0.25 Hz range can be consistently characterized by four intrinsic frequency clusters, centered at 0.028 Hz (IMF4), 0.080 Hz (IMF3), 0.15 Hz (IMF2) and 0.22 Hz (IMF1); (2) these frequency clusters were highly reproducible, and independent of rs-fMRI data sampling rate; (3) not all frequencies were associated with equivalent network topology, in contrast to previous findings. In fact, while IMF4 is most likely associated with physiological fluctuations due to respiration and pulse, IMF3 is most likely associated with metabolic processes, and IMF2 with vasomotor activity. Both IMF3 and IMF4 could produce the brain-network topology typically observed in fMRI, whereas IMF1 and IMF2 could not. These findings provide initial evidence of feasibility in decomposing the rs-fMRI signal into its intrinsic oscillatory frequencies in a reproducible manner.

Spatiotemporal Empirical Mode Decomposition of Resting-State fMRI Signals: Application to Global Signal Regression.

Resting-state functional connectivity MRI (rs-fcMRI) is a common method for mapping functional brain networks. However, estimation of these networks is affected by the presence of a common global systemic noise, or global signal (GS). Previous studies have shown that the common preprocessing steps of removing the GS may create spurious correlations between brain regions. In this paper, we decompose fMRI signals into 5 spatial and 3 temporal intrinsic mode functions (SIMF and TIMF, respectively) by means of the empirical mode decomposition (EMD), which is an adaptive data-driven method widely used to analyze non-linear and non-stationary phenomena. For each SIMF, functional connectivity matrices were computed by means of Pearson correlation between TIMFs of different brain areas. Thus, instead of a single connectivity matrix, we obtained 5 × 3 = 15 functional connectivity matrices. Given the high correlation and global efficiency values of the connectivity matrices related to the low spatial maps (SIMF3, SIMF4, and SIMF5), our results suggest that these maps can be considered as spatial global signal masks. Thus, by summing up the first two SIMFs extracted from the fMRI signals, we have automatically excluded the GS which is now voxel-specific. We compared the performance of our method with the conventional GS regression and to the results when the GS was not removed. While the correlation pattern identified by the other methods suffers from a low level of precision in identifying the correct brain network connectivity, our approach demonstrated expected connectivity patterns for the default mode network and task-positive network.

A Strategy for Classification of "Vaginal vs. Cesarean Section" Delivery: Bivariate Empirical Mode Decomposition of Cardiotocographic Recordings.

We propose objective and robust measures for the purpose of classification of "vaginal vs. cesarean section" delivery by investigating temporal dynamics and complex interactions between fetal heart rate (FHR) and maternal uterine contraction (UC) recordings from cardiotocographic (CTG) traces. Multivariate extension of empirical mode decomposition (EMD) yields intrinsic scales embedded in UC-FHR recordings while also retaining inter-channel (UC-FHR) coupling at multiple scales. The mode alignment property of EMD results in the matched signal decomposition, in terms of frequency content, which paves the way for the selection of robust and objective time-frequency features for the problem at hand. Specifically, instantaneous amplitude and instantaneous frequency of multivariate intrinsic mode functions are utilized to construct a class of features which capture nonlinear and nonstationary interactions from UC-FHR recordings. The proposed features are fed to a variety of modern machine learning classifiers (decision tree, support vector machine, AdaBoost) to delineate vaginal and cesarean dynamics. We evaluate the performance of different classifiers on a real world dataset by investigating the following classifying measures: sensitivity, specificity, area under the ROC curve (AUC) and mean squared error (MSE). It is observed that under the application of all proposed 40 features AdaBoost classifier provides the best accuracy of 91.8% sensitivity, 95.5% specificity, 98% AUC, and 5% MSE. To conclude, the utilization of all proposed time-frequency features as input to machine learning classifiers can benefit clinical obstetric practitioners through a robust and automatic approach for the classification of fetus dynamics.

An optimized design of seizure detection system using joint feature extraction of multichannel EEG signals.

The detection of seizure onset and events using electroencephalogram (EEG) signals are important tasks in epilepsy research. The literature available on seizure detection has discussed the implementation of advanced signal processing algorithms using tools accessed over the cloud. However, seizure monitoring application needs near sensor processing due to privacy and latency issues. In this paper, a real time seizure detection system has been implemented using an embedded system. The proposed system is based on ensemble empirical mode decomposition (EEMD) and tunable-Q wavelet transform (TQWT) algorithms. The analysis and classification of non-stationary EEG signals require the wavelet transform with high Q-factor. However, direct use of TQWT increases the computational complexity of feature extraction from multivariate EEG signals. In this paper, the first step is to process the signal by using EEMD to obtain 8 intrinsic mode functions (IMFs). The Kraskov (KraEn), sample (SampEn), and permutation (PermEn) entropy features of IMFs are extracted and based on optimum values, and 4 IMFs are decomposed using TQWT. Secondly, centered correntropy (CenCorrEn) features of the 1 st and 16 th sub-band of TQWT have been used as classifier inputs. The performance of multilayer perceptron neural networks (MLPNN), least squares support vector machine (LSSVM), and random forest (RF) classifiers has been tested on the multichannel EEG data recorded from a local hospital. The RF classifier has produced the highest accuracy of 96.2% in classifying the signals. The proposed scheme has been employed in developing an embedded seizure detection system to assist neurologists in making seizure diagnostic decisions.

Advances in functional magnetic resonance imaging data analysis methods using Empirical Mode Decomposition to investigate temporal changes in early Parkinson's disease.

INTRODUCTION: Previous neuroimaging studies of Parkinson's disease (PD) patients have shown changes in whole-brain functional connectivity networks. Whether connectivity changes can be detected in the early stages (first 3 years) of PD by resting-state functional magnetic resonance imaging (fMRI) remains elusive. Research infrastructure including MRI and analytic capabilities is required to investigate this issue. The National Institutes of Health/National Institute of General Medical Sciences Center for Biomedical Research Excellence awards support infrastructure to advance research goals. METHODS: Static and dynamic functional connectivity analyses were conducted on early stage never-medicated PD subjects (N = 18) and matched healthy controls (N = 18) from the Parkinson's Progression Markers Initiative. RESULTS: Altered static and altered dynamic functional connectivity patterns were found in early PD resting-state fMRI data. Most static networks (with the exception of the default mode network) had a reduction in frequency and energy in specific low-frequency bands. Changes in dynamic networks in PD were associated with a decreased switching rate of brain states. DISCUSSION: This study demonstrates that in early PD, resting-state fMRI networks show spatial and temporal differences of fMRI signal characteristics. However, the default mode network was not associated with any measurable changes. Furthermore, by incorporating an optimum window size in a dynamic functional connectivity analysis, we found altered whole-brain temporal features in early PD, showing that PD subjects spend significantly more time than healthy controls in a specific brain state. These findings may help in improving diagnosis of early never-medicated PD patients. These key observations emerged in a Center for Biomedical Research Excellence-supported research environment.

Rub-Impact Fault Diagnosis Using an Effective IMF Selection Technique in Ensemble Empirical Mode Decomposition and Hybrid Feature Models.

The complex nature of rubbing faults makes it difficult to use traditional signal analysis methods for feature extraction. Various time-frequency analysis approaches based on signal decomposition, such as empirical mode decomposition (EMD) and ensemble EMD (EEMD), have been widely utilized recently to analyze rub-impact faults. However, traditional EMD suffers from “mode-mixing”, and in both EMD and EEMD the relevance of the extracted components to rubbing processes must be determined. In this paper, we introduce a new informative intrinsic mode function (IMF) selection method for EEMD and a hybrid feature model for diagnosing rub-impact faults of various intensities. Our method uses a novel selection procedure that combines the degree-of-presence ratio of rub impact and a Kullback⁻Leibler divergence-based similarity measure into an IMF quality metric with adaptive threshold-based selection to pick the meaningful signal-dominant modes. Signals reconstructed using the selected IMFs contained explicit information about the rubbing faults and are used for hybrid feature extraction. Experimental results demonstrated that the proposed approach effectively defines meaningful IMFs for rubbing processes, and the presented hybrid feature model allows for the classification of rub-impact faults of various intensities with good accuracy.

Relevant Features Selection for Automatic Prediction of Preterm Deliveries from Pregnancy ElectroHysterograhic (EHG) records.

In this study, we proposed an approach able to predict whether a pregnant woman with contractions would give birth earlier than expected (i.e., before the 37 t h week of gestation (WG)). It only processes non-invasive electrohysterographic (EHG) signals fully automatically without assistance of an expert or an additional medical system. We used term and preterm EHG signals of 30-minutes duration collected between the 27 t h and the 32 n d WG. Preterm deliveries (< 37W G) had occurred in average 4.00 ± 1.88 weeks since recording dates. Each recording contains three bipolar channels. Using the Huang-Hilbert transform (HHT), we obtained up to twelve intrinsic mode functions (IMFs) for each signal. We found that the most relevant IMFs for both term and preterm records were IMF3 and IMF6. From these two IMFs, we extracted 8 most relevant features targeting EHG signal specificities. We investigated features classifications using support vector machine (SVM) for the 3 single-channels and for all their possible combinations. High discrimination power between term and preterm EHG records was obtained with linear-SVM classifiers. For almost all the cases, mean areas under curves (AUC) exceeded 0.92. A two-channel combination (7 features) achieved the best mean results with A c c u r a c y = 95.70%, S e n s i t i v i t y = 98.40%, S p e c i f i c i t y = 93.00% and A U C = 0.95. Results of the three-channel combination (9 features) were A c c u r a c y = 92.30%, S e n s i t i v i t y = 93.00%, S p e c i f i c i t y = 91.60% and A U C = 0.96. The best single-channel (8 features) gave the mean values: A c c u r a c y = 90.40%, S e n s i t i v i t y = 93.60% and A U C = 0.94. Thus, the advantage of our approach is the high diagnostic performance at low computational cost.