Signal Segments Research Articles

Classification of Blood Pressure Levels Based on Photoplethysmogram and Electrocardiogram Signals with a Concatenated Convolutional Neural Network.

Hypertension is a severe public health issue worldwide that significantly increases the risk of cardiac vascular disease, stroke, brain hemorrhage, and renal dysfunction. Early screening of blood pressure (BP) levels is essential to prevent the dangerous complication associated with hypertension as the leading cause of death. Recent studies have focused on employing photoplethysmograms (PPG) with machine learning to classify BP levels. However, several studies claimed that electrocardiograms (ECG) also strongly correlate with blood pressure. Therefore, we proposed a concatenated convolutional neural network which integrated the features extracted from PPG and ECG signals. This study used the MIMIC III dataset, which provided PPG, ECG, and arterial blood pressure (ABP) signals. A total of 14,298 signal segments were obtained from 221 patients, which were divided into 9150 signals of train data, 2288 signals of validation data, and 2860 signals of test data. In the training process, five-fold cross-validation was applied to select the best model with the highest classification performance. The proposed concatenated CNN architecture using PPG and ECG obtained the highest test accuracy of 94.56-95.15% with a 95% confidence interval in classifying BP levels into hypotension, normotension, prehypertension, hypertension stage 1, and hypertension stage 2. The result shows that the proposed method is a promising solution to categorize BP levels effectively, assisting medical personnel in making a clinical diagnosis.

A biosensing system employing nonlinear dynamic analysis-assisted neural network for drug-induced cardiotoxicity assessment

Preclinical investigation of drug-induced cardiotoxicity is of importance for drug development. To evaluate such cardiotoxicity, in vitro high-throughput interdigitated electrode-based recording of cardiomyocytes mechanical beating is widely used. To automatically analyze the features from the beating signals for drug-induced cardiotoxicity assessment, artificial neural network analysis is conventionally employed and signals are segmented into cycles and feature points are located in the cycles. However, signal segmentation and location of feature points for different signal shapes require design of specific algorithms. Consequently, this may lower the efficiency of research and the applications of such algorithms in signals with different morphologies are limited. Here, we present a biosensing system that employs nonlinear dynamic analysis-assisted neural network (NDANN) to avoid the signal segmentation process and directly extract features from beating signal time series. By processing beating time series with fixed time duration to avoid the signal segmentation process, this NDANN-based biosensing system can identify drug-induced cardiotoxicity with accuracy over 0.99. The individual drugs were classified with high accuracies over 0.94 and drug-induced cardiotoxicity levels were accurately predicted. We also evaluated the generalization performance of the NDANN-based biosensing system in assessing drug-induced cardiotoxicity through an independent dataset. This system achieved accuracy of 0.85–0.95 for different drug concentrations in identification of drug-induced cardiotoxicity. This result demonstrates that our NDANN-based biosensing system has the capacity of screening newly developed drugs, which is crucial in practical applications. This NDANN-based biosensing system can work as a new screening platform for drug-induced cardiotoxicity and improve the efficiency of bio-signal processing.

Evaluation of Hepatic Fibrosis Using Ultrasound Backscattered Radiofrequency Signals and One-Dimensional Convolutional Neural Networks.

The early detection of hepatic fibrosis is of critical importance. Ultrasound backscattered radiofrequency signals from the liver contain abundant information about its microstructure. We proposed a method for characterizing human hepatic fibrosis using one-dimensional convolutional neural networks (CNNs) based on ultrasound backscattered signals. The proposed CNN model was composed of four one-dimensional convolutional layers, four one-dimensional max-pooling layers, and four fully connected layers. Ultrasound radiofrequency signals collected from 230 participants (F0: 23; F1: 46; F2: 51; F3: 49; F4: 61) with a 3-MHz transducer were analyzed. Liver regions of interest (ROIs) that contained most of the liver ultrasound backscattered signals were manually delineated using B-mode images reconstructed from the backscattered signals. ROI signals were normalized and augmented by using a sliding window technique. After data augmentation, the radiofrequency signal segments were divided into training sets, validation sets and test sets at a ratio of 80%:10%:10%. In the test sets, the proposed algorithm produced an area under the receive operating characteristic curve of 0.933 (accuracy: 91.30%; sensitivity: 92.00%; specificity: 90.48%), 0.997 (accuracy: 94.29%; sensitivity: 94.74%; specificity: 93.75%), 0.818 (accuracy: 75.00%; sensitivity: 69.23%; specificity: 81.82%), and 0.934 (accuracy: 91.67%; sensitivity: 88.89%; specificity: 94.44%) for diagnosis liver fibrosis stage ≥F1, ≥F2, ≥F3, and ≥F4, respectively. Experimental results indicated that the proposed deep learning algorithm based on ultrasound backscattered signals yields a satisfying performance when diagnosing hepatic fibrosis stages. The proposed method may be used as a new quantitative ultrasound approach to characterizing hepatic fibrosis.

Analytical non‐Fourier solution of linear time‐invariant circuits with periodic inputs

SummaryA systematic approach for solving linear time‐invariant electric circuits with periodic input signals is presented. The method involves the following steps: (1) The governing differential equation for the desired output is solved for each time segment of the input signal in one period, leaving the constant coefficients of the homogeneous parts of the solution as unknowns. For an nth‐order circuit with a source consisting of q segments in one period, there are such unknown coefficients. (2) Using the solution obtained in step 1, the voltage of each capacitor and the current of each inductor are determined. (3) Transition conditions for the voltages and currents found in step 2 at the transition points of successive segments and also periodicity conditions for the beginning and end points of the period are applied. (4) Implementation of step 3 results in a system of equations in terms of the unknown coefficients described in step 1. Solving this system of equations, the circuit response in one period becomes readily available.

Signal Quality Index for the fetal heart rates: Development and improvements for fetal monitoring

Fetal heart rate (FHR) monitoring is the most widely used tool in clinics to assess fetal health. However, FHR with low-quality signals may somehow exaggerate the risk of the fetus suffering from acidemia, thus contributing to an increase in cesarean section rates. If there is an algorithm capable to identify and reject low-quality signals and then calculate heart rate parameters only on high-quality signals, inappropriate obstetric interventions may be greatly reduced. Currently, only the Signal Quality Index (SQI) is available for adults. When we applied SQI to fetal data, its performance drops dramatically. To fill the gap in the fetal signal quality index, we developed an easy-to-use Fetal heart rate Signal Quality Index (FSQI) in this paper. Firstly, we used a human-in-the-loop strategy to reduce the amount of data annotation effort. Secondly, we developed the FSQI algorithm oriented towards the fetus to identify low-quality signal segments. Further, we proposed a post-processing technique based on fine-grained recognition to identify low-quality signal segments more accurately. Finally, we applied the FSQI algorithm in combination with the existing FHR deceleration event detection algorithm to test the effectiveness of the algorithm. The results showed that the FSQI algorithm we developed achieved an overall accuracy of 99.8842% in signal quality classification, while also eliminating 94.92% of the incorrectly detected deceleration events of current state-of-the-art methods.

Vital_sqi: A Python package for physiological signal quality control.

Electrocardiogram (ECG) and photoplethysmogram (PPG) are commonly used to determine the vital signs of heart rate, respiratory rate, and oxygen saturation in patient monitoring. In addition to simple observation of those summarized indexes, waveform signals can be analyzed to provide deeper insights into disease pathophysiology and support clinical decisions. Such data, generated from continuous patient monitoring from both conventional bedside and low-cost wearable monitors, are increasingly accessible. However, the recorded waveforms suffer from considerable noise and artifacts and, hence, are not necessarily used prior to certain quality control (QC) measures, especially by those with limited programming experience. Various signal quality indices (SQIs) have been proposed to indicate signal quality. To facilitate and harmonize a wider usage of SQIs in practice, we present a Python package, named vital_sqi, which provides a unified interface to the state-of-the-art SQIs for ECG and PPG signals. The vital_sqi package provides with seven different peak detectors and access to more than 70 SQIs by using different settings. The vital_sqi package is designed with pipelines and graphical user interfaces to enable users of various programming fluency to use the package. Multiple SQI extraction pipelines can take the PPG and ECG waveforms and generate a bespoke SQI table. As these SQI scores represent the signal features, they can be input in any quality classifier. The package provides functions to build simple rule-based decision systems for signal segment quality classification using user-defined SQI thresholds. An experiment with a carefully annotated PPG dataset suggests thresholds for relevant PPG SQIs.

Power normalized cepstral robust features of deep neural networks in a cloud computing data privacy protection scheme

Deep Neural Networks (DNNs) have developed rapidly in data privacy protection applications such as medical treatment and finance. However, DNNs require high-speed and high-memory computers in terms of computation, otherwise training can be very lengthy. Furthermore, DNNs are often not available in resource-constrained mobile devices. Therefore, training and executing DNNs are increasingly using cloud computing. In the paper, the Power Normalized Cepstrum-based Robust Feature Detector (PNC-RFD), with deep learning in the cloud computing, is proposed for data privacy protection. The proposed PNC-RFD extracts a specified number of signal segments of high robustness used to embed and extract various data. For the sake of embedding and extracting the data, a method of information hiding employing Dual-Tree Complex Wavelet Packet Transform (DT CWPT) is therefore presented. The presented scheme simultaneously embeds multiple data into coefficients of the DT CWPT of signal segments. By embedding the data into the orthogonal spaces, the proposed method ensures the independent extraction of the multiple data. In line with the performance analysis, the superiority of the presented scheme is elaborated through making the comparison with the current state-of-the-art methods.

Optimal Driving Strategies for a Fleet of Trains on Level Track with Prescribed Intermediate Signal Times and Safe Separation

We propose an analytic solution to the problem of finding optimal driving strategies that minimize total tractive energy consumption for a fleet of trains traveling on the same track in the same direction subject to clearance-time equality constraints that ensure safe separation and compress the headway timespan. We assume the track is divided into sections by a set of trackside signals at fixed positions. For each intermediate signal there is an associated signal segment consisting of the two adjacent sections. Successive trains are safely separated only if the leading train leaves the signal segment before the following train enters. Although the fleet can be safely separated by a complete set of clearance times and associated clearance-time inequality constraints the problem of finding optimal schedules with safe separation rapidly becomes intractable as the number of trains and signals increases. The main difficulty is in distinguishing between active equality constraints and inactive inequality constraints. The curse of dimensionality means it is not feasible to check every different combination of active constraints, find the optimal strategies for each train, optimize the corresponding prescribed times and calculate the cost. Nevertheless we can formulate and solve an alternative problem with active clearance-time equality constraints for successive trains defined at selected signals. We show that this problem can be formulated as an unconstrained convex optimization and propose a solution algorithm that finds the optimal schedule and the associated optimal strategies for each train. Finally we find optimal schedules for a case study using realistic parameters on a busy metropolitan line.

A robust quantitative approach for laser speckle contrast imaging perfusion analysis revealed anomalies in the brain blood flow in offspring mice of preeclampsia.

Microcirculation analysis of the brain cortex is challenging because surface perfusion varies rapidly in small space-time regions and is bone protected. The laser speckle contrast imaging (LSCI) technique allows analyzing in vivo brain vascular perfusion generating a large amount of data that requires sophisticated data analytics, making researchers invest much effort in processing. Our research question was whether the reduced placental perfusion model (RUPP) of preeclampsia (PE) was associated with impaired blood perfusion in the offspring's brains. We aimed to develop a robust numerical approach that mainly consisted of applying a signal-processing tool for calculating optimal segmentation and piece-wise fits of the offspring's brain perfusion signals obtained from the LSCI technique. We combined this tool with the usual statistical analysis, implementing both in Matlab software. We performed brain perfusion measurements from offspring (five days postnatal, P5) of control pregnant dams (sham, n=13) and of RUPP dams (RUPP, n=7) using the Pericam® PSI-HR system at a basal condition and after thermal stimuli (warm and cold). We found that pups of RUPP mice exhibited significant differences in perfusion and vascular response to thermal stimuli compared to the sham mice. These differences were associated with high data variability in the Sham group, while in the RUPP group, perfusion looks "stiffer." Data also suggest sex-dimorphism in the vascular response since female pups in the Sham group but not male pups showed statistically significant differences in response to the warm stimulus. Again, this sex-related difference was absent in pups of RUPP mice. In conclusion, we present a robust quantitative approach for LSCI measurements that revealed anomalies in the brain blood flow in offspring of the RUPP model of PE.

CSI-Ratio-Based Doppler Frequency Estimation in Integrated Sensing and Communications

Estimating the Doppler frequency is an important part of sensing moving targets in integrated sensing and communications (ISAC) systems, such as human tracking and activity recognition. However, it can be highly challenging when there is clock asynchronism between the transmitter (Tx) and the receiver (Rx), in bistatic setups that are common nowadays. In this article, we propose three algorithms for Doppler frequency estimation based on the ratio of channel state information (CSI). These algorithms explore different properties of the CSI ratio, including the circle-preserving property of the Mobius transform, the periodicity of the CSI ratio, and the difference (or correlation) between segments of CSI-ratio signals. Experimental results demonstrate that the proposed algorithms can estimate Doppler frequency accurately, outperforming the commonly used approach based on cross-antenna cross correlation (CACC).

Erroneous Vehicle Velocity Estimation Correction Using Anisotropic Magnetoresistive (AMR) Sensors.

Magnetic field sensors installed in the road infrastructure can be used for autonomous traffic flow parametrization. Although the main goal of such a measuring system is the recognition of the class of vehicle and classification, velocity is the essential parameter for further calculation and it must be estimated with high reliability. In-field test campaigns, during actual traffic conditions, showed that commonly accepted velocity estimation methods occasionally produce highly erroneous results. For anomaly detection, we propose a criterion and two different correction algorithms. Non-linear signal rescaling and time-based segmentation algorithms are presented and compared for faulty result mitigation. The first one consists of suppressing the highly distorted signal peaks and looking for the best match with cross-correlation. The second approach relies on signals segmentation according to the feature points and multiple cross-correlation comparisons. The proposed two algorithms are evaluated with a dataset of over 300 magnetic signatures of a vehicle from unconstraint traffic conditions. Results show that the proposed criteria highlight all greatly faulty results and that the correction algorithms reduce the maximum error by twofold, but due to the increased mean error, mitigation technics shall be used explicitly with distorted signals.

Research on the Effectiveness of Deep Convolutional Neural Network for Electromagnetic Interference Identification Based on I/Q Data

With the development of wireless communication technology, the electromagnetic interference (EMI) of artificial radio to weather radar increases significantly, which has a serious impact on the quality of radar data. Most of the research on detecting and suppressing electromagnetic interference was based on the primary product of the radar. This paper researches the effectiveness of deep convolutional neural networks (DCNN) to identify and suppress electromagnetic interference based on the I/Q data output from the front end of a radar receiver. Firstly, this paper selected UNet, ResNet with UNet structure, and DeepLab V3+ for the semantic segmentation of electromagnetic interference and other signals. After semantic segmentation, this paper used the linear interpolation method to suppress EMI. Finally, this paper selected the prediction precision of the model and compared the quality of primary products before and after EMI suppression to evaluate the effectiveness of DCNN. The results showed that all three models could effectively identify the electromagnetic interference and the quality of the data were improved after suppression. It suggests that the use of DCNN on the I/Q data output from the front end of a radar receiver can play a certain effect on the identification of electromagnetic interference.

Recognition Method for Broiler Sound Signals Based on Multi-Domain Sound Features and Classification Model.

In view of the limited number of extracted sound features, the lack of in-depth analysis of applicable sound features, and the lack of in-depth study of the selection basis and optimization process of classification models in the existing broiler sound classification or recognition research, the author proposes a recognition method for broiler sound signals based on multi-domain sound features and classification models. The implementation process is divided into the training stage and the testing stage. In the training stage, the experimental area is built, and multiple segments of broiler sound signals are collected and filtered. Through sub-frame processing and endpoint detection, the combinations of start frames and end frames of multiple sound types in broiler sound signals are obtained. A total of sixty sound features from four aspects of time domain, frequency domain, Mel-Frequency Cepstral Coefficients (MFCC), and sparse representation are extracted from each frame signal to form multiple feature vectors. These feature vectors are labeled manually to build the data set. The min-max standardization method is used to process the data set, and the random forest is used to calculate the importance of sound features. Then, thirty sound features that contribute more to the classification effect of the classification model are retained. On this basis, the classification models based on seven classification algorithms are trained, the best-performing classification model based on k-Nearest Neighbor (kNN) is obtained, and its inherent parameters are optimized. Then, the optimal classification model is obtained. The test results show that the average classification accuracy achieved by the decision-tree-based classifier (abbreviated as DT classifier) on the data set before and after min–max standardization processing is improved by 0.6%, the average classification accuracy achieved by the DT classifier on the data set before and after feature selection is improved by 3.1%, the average classification accuracy achieved by the kNN-based classification model before and after parameter optimization is improved by 1.2%, and the highest classification accuracy is 94.16%. In the testing stage, for a segment of the broiler sound signal collected in the broiler captivity area, the combinations of the start frames and end frames of multiple sound types in the broiler sound signal are obtained through signal filtering, sub-frame processing, endpoint detection, and other steps. Thirty sound features are extracted from each frame signal to form the data set to be predicted. The optimal classification model is used to predict the labels of each piece of data in the data set to be predicted. By performing majority voting processing on the predicted labels of the data combination corresponding to each sound type, the common labels are obtained; that is, the predicted types are obtained. On this basis, the definition of recognition accuracy for broiler sound signals is proposed. The test results show that the classification accuracy achieved by the optimal classification model on the data set to be predicted is 93.57%, and the recognition accuracy achieved on the multiple segments of the broiler sound signals is 99.12%.

Prediagnosis of Heart Failure (HF) Using Deep Learning and the Korotkoff Sound

Heart failure (HF) is a devastating condition that impairs people’s lives and health. Because of the high morbidity and mortality associated with HF, early detection is becoming increasingly critical. Many studies have focused on the field of heart disease diagnosis based on heart sound (HS), demonstrating the feasibility of sound signals in heart disease diagnosis. In this paper, we propose a non-invasive early diagnosis method for HF based on a deep learning (DL) network and the Korotkoff sound (KS). The accuracy of the KS-based HF prediagnosis was investigated utilizing continuous wavelet transform (CWT) features, Mel frequency cepstrum coefficient (MFCC) features, and signal segmentation. Fivefold cross-validation was applied to the four DL models: AlexNet, VGG19, ResNet50, and Xception, and the performance of each model was evaluated using accuracy (Acc), specificity (Sp), sensitivity (Se), area under curve (AUC), and time consumption (Tc). The results reveal that the performance of the four models on MFCC datasets is significantly improved when compared to CWT datasets, and each model performed considerably better on the non-segmented dataset than on the segmented dataset, indicating that KS signal segmentation and feature extraction had a significant impact on the KS-based CHF prediagnosis performance. Our method eventually achieves the prediagnosis results of Acc (96.0%), Se (97.5%), and Sp (93.8%) based on a comparative study of the model and the data set. The research demonstrates that the KS-based prediagnosis method proposed in this paper could accomplish accurate HF prediagnosis, which will offer new research approaches and a more convenient way to achieve early HF prevention.

An Assessment of Waveform Processing for a Single-Beam Bathymetric LiDAR System (SBLS-1).

The single-beam bathymetric light detection and ranging (LiDAR) system 1 (SBLS-1), which is equipped with a 532-nm-band laser projector and two concentric-circle receivers for shallow- and deep-water echo signals, is a lightweight and convenient prototype instrument with low energy consumption. In this study, a novel LiDAR bathymetric method is utilized to achieve single-beam and dual-channel bathymetric characteristics, and an adaptive extraction method is proposed based on the cumulative standard deviation of the peak and trough, which is mainly used to extract the signal segment and eliminate system and random noise. To adapt the dual-channel bathymetric mechanism, an automatic channel-selection method was used at various water depths. A minimum half-wavelength Gaussian iterative decomposition is proposed to improve the detection accuracy of the surface- and bottom-water waveform components and ensure bathymetric accuracy and reliability. Based on a comparison between the experimental results and in situ data, it was found that the SBLS-1 obtained a bathymetric accuracy and RMSE of 0.27 m and 0.23 m at the Weifang and Qingdao test fields. This indicates that the SBLS-1 was bathymetrically capable of acquiring a reliable, high-efficiency waveform dataset. Hence, the novel LiDAR bathymetric method can effectively achieve high-accuracy near-shore bathymetry.

Estimating steelpan note class from attack transients

The estimation of fundamental frequency of instruments is an important task in computational audio analysis. The current state of the art methods use neural networks for this task. This process is typically computed periodically over very short segments of a monophonic audio signal so that minute shifts in intonation can be detected. However, the steelpan has discretely tuned notes where the performer has no direct control over pitch once a note has been activated. The activation of a note has great influence over the acoustical properties of the resultant note. Much research has been devoted to the tonality, construction, and acoustical properties of steelpans, but relatively little focuses on the attack transient specifically.This paper evaluates the application of pitch detection methods to the attack transients of steelpan notes. A dataset containing labeled audio samples from multiple tenor steelpans is used for training and evaluation. The accuracy of this approach for pitch detection is compared with established methods applied to both entire notes and only attack transients. Determining a steelpan note’s pitch from the attack transient is an important first step in building a robust low latency automatic transcription system that can be used for both analysis as well as live performance.

Electromyography signal segmentation method based on spectral subtraction backtracking

Electromyography signal segmentation method based on spectral subtraction backtracking

A Novel Supervised Filter Feature Selection Method Based on Gaussian Probability Density for Fault Diagnosis of Permanent Magnet DC Motors.

For permanent magnet DC motors (PMDCMs), the amplitude of the current signals gradually decreases after the motor starts. In this work, the time domain features and time-frequency-domain features extracted from several successive segments of current signals make up a feature vector, which is adopted for fault diagnosis of PMDCMs. Many redundant features will lead to a decrease in diagnosis efficiency and increase the computation cost, so it is necessary to eliminate redundant features and features that have negative effects. This paper presents a novel supervised filter feature selection method for reducing data dimension by employing the Gaussian probability density function (GPDF) and named Gaussian vote feature selection (GVFS). To evaluate the effectiveness of the proposed GVFS, we compared it with the other five filter feature selection methods by utilizing the PMDCM’s data. Additionally, Gaussian naive Bayes (GNB), -nearest neighbor algorithm (-NN), and support vector machine (SVM) are utilized for the construction of fault diagnosis models. Experimental results show that the proposed GVFS has a better diagnostic effect than the other five feature selection methods, and the average accuracy of fault diagnosis improves from 97.89% to 99.44%. This paper lays the foundation of fault diagnosis for PMDCMs and provides a novel filter feature selection method.

A compact hybrid fully convolutional and BiLSTM network with squeeze and temporal excitation approach for semantic segmentation of ECG signal

A compact hybrid fully convolutional and BiLSTM network with squeeze and temporal excitation approach for semantic segmentation of ECG signal

Materials property mapping from atomic scale imaging via machine learning based sub-pixel processing

Direct visualization of the atomic structure in scanning transmission electron microscopy has led to a comprehensive understanding of the structure-property relationship. However, a reliable characterization of the structural transition on a picometric scale is still challenging because of the limited spatial resolution and noise. Here, we demonstrate that the primary segmentation of atomic signals from background, succeeded by a denoising process, enables structural analysis in a sub-pixel accuracy. Poisson noise is eliminated using the block matching and three-dimensional filtering with Anscombe transformation, and remnant noise is removed via morphological filtering, which results in an increase of peak signal-to-noise ratio from 7 to 11 dB. Extracting the centroids of atomic columns segmented via K-means clustering, an unsupervised method for robust thresholding, achieves an average error of less than 0.7 pixel, which corresponds to 4.6 pm. This study will contribute to a profound understanding of the local structural dynamics in crystal structures.