NEURAL NETWORK AUTOENCODER MODEL FOR FORMING REDUCED VECTOR CHARACTERISTICS OF ECG SIGNALS

The study intended to explore the effect of different neural network algorithms in the electrocardiogram (ECG) classification of patients with congenital heart disease (CHD). Based on the single convolutional neural network (CNN) ECG algorithm and the recurrent neural network (RNN) ECG algorithm, a multimodal neural network (MNN) ECG algorithm was constructed utilizing the MIT-BIH database as training set and test set. Furthermore, the MNN ECG algorithm was optimized to establish an improved MNN (IMNN) algorithm, which was applied to the diagnosis of CHD patients. The CHD patients admitted between August 2016 and August 2019 were selected for analysis to compare the classification effect and accuracy rate of IMNN, MNN, CNN ECG, and RNN ECG algorithms. It was found that the RNN ECG algorithm had higher classification sensitivity and true positive rate in terms of normal or bundle (NB) branch block beat, supraventricular abnormal (SA) rhythm, abnormal ventricular (AV) beat, and fusion beat (FB) than the CNN ECG algorithm ( P < 0.05 ), and the classification sensitivity and true positive rate of IMNN algorithm in the four aspects were significantly higher than those of MNN algorithm ( P < 0.05 ). The classification accuracy of CNN ECG algorithm and RNN ECG algorithm was above 98%, while that of MNN algorithm and IMNN algorithm was better than that of CNN ECG algorithm and RNN ECG algorithm, and the accuracy rate can reach 98.5% or more. Moreover, the accuracy rate of the IMNN algorithm can reach more than 98%. In conclusion, IMNN not only has a good classification ability in the simulated environment but also performs well in the actual environment, which is worthy of clinical promotion.

In many medical applications, feature selection is obvious; but in medical domains, selecting features and creating a feature vector may require more effort. The wavelet transform (WT) technique is used to identify the characteristic points of an electrocardiogram (ECG) signal with fairly good accuracy, even in the presence of severe high-frequency and low-frequency noise. Principal component analysis (PCA) is a suitable technique for ECG data analysis, feature extraction, and image processing — an important technique that is not based upon a probability model. The aim of the paper is to derive better diagnostic parameters for reducing the size of ECG data while preserving morphology, which can be done by PCA. In this analysis, PCA is used for decorrelation of ECG signals, noise, and artifacts from various raw ECG data sets. The aim of this paper is twofold: first, to describe an elegant algorithm that uses WT alone to identify the characteristic points of an ECG signal; and second, to use a composite WT-based PCA method for redundant data reduction and better feature extraction. PCA scatter plots can be observed as a good basis for feature selection to account for cardiac abnormalities. The study is analyzed with higher-order statistics, in contrast to the conventional methods that use only geometric characteristics of feature waves and lower-order statistics. A new algorithm — viz. PCA variance estimator — is developed for this analysis, and the results are also obtained for different combinations of leads to find correlations for feature classification and useful diagnostic information. PCA scatter plots of various chest and augmented ECG leads are obtained to examine the varying orientations of the ECG data in different quadrants, indicating the cardiac events and abnormalities. The efficacy of the PCA algorithm is tested on different leads of 12-channel ECG data; file no. 01 of the Common Standards for Electrocardiography (CSE) database is used for this study. Better feature extraction is obtained for some specific combinations of leads, and significant improvement in signal quality is achieved by identifying the noise and artifact components. The quadrant analysis discussed in this paper highlights the filtering requirements for further ECG processing after performing PCA, as a primary step for decorrelation and dimensionality reduction. The values of the parameters obtained from the results of PCA are also compared with those of wavelet methods.

Millions of clinicians rely daily on automated preliminary electrocardiogram (ECG) interpretation. Critical comparisons of machine learning-based automated analysis against clinically accepted standards of care are lacking. To use readily available 12-lead ECG data to train and apply an explainability technique to a convolutional neural network (CNN) that achieves high performance against clinical standards of care. This cross-sectional study was conducted using data from January 1, 2003, to December 31, 2018. Data were obtained in a commonly available 12-lead ECG format from a single-center tertiary care institution. All patients aged 18 years or older who received ECGs at the University of California, San Francisco, were included, yielding a total of 365 009 patients. Data were analyzed from January 1, 2019, to March 2, 2021. A CNN was trained to predict the presence of 38 diagnostic classes in 5 categories from 12-lead ECG data. A CNN explainability technique called LIME (Linear Interpretable Model-Agnostic Explanations) was used to visualize ECG segments contributing to CNN diagnoses. Area under the receiver operating characteristic curve (AUC), sensitivity, and specificity were calculated for the CNN in the holdout test data set against cardiologist clinical diagnoses. For a second validation, 3 electrophysiologists provided consensus committee diagnoses against which the CNN, cardiologist clinical diagnosis, and MUSE (GE Healthcare) automated analysis performance was compared using the F1 score; AUC, sensitivity, and specificity were also calculated for the CNN against the consensus committee. A total of 992 748 ECGs from 365 009 adult patients (mean [SD] age, 56.2 [17.6] years; 183 600 women [50.3%]; and 175 277 White patients [48.0%]) were included in the analysis. In 91 440 test data set ECGs, the CNN demonstrated an AUC of at least 0.960 for 32 of 38 classes (84.2%). Against the consensus committee diagnoses, the CNN had higher frequency-weighted mean F1 scores than both cardiologists and MUSE in all 5 categories (CNN frequency-weighted F1 score for rhythm, 0.812; conduction, 0.729; chamber diagnosis, 0.598; infarct, 0.674; and other diagnosis, 0.875). For 32 of 38 classes (84.2%), the CNN had AUCs of at least 0.910 and demonstrated comparable F1 scores and higher sensitivity than cardiologists, except for atrial fibrillation (CNN F1 score, 0.847 vs cardiologist F1 score, 0.881), junctional rhythm (0.526 vs 0.727), premature ventricular complex (0.786 vs 0.800), and Wolff-Parkinson-White (0.800 vs 0.842). Compared with MUSE, the CNN had higher F1 scores for all classes except supraventricular tachycardia (CNN F1 score, 0.696 vs MUSE F1 score, 0.714). The LIME technique highlighted physiologically relevant ECG segments. The results of this cross-sectional study suggest that readily available ECG data can be used to train a CNN algorithm to achieve comparable performance to clinical cardiologists and exceed the performance of MUSE automated analysis for most diagnoses, with some exceptions. The LIME explainability technique applied to CNNs highlights physiologically relevant ECG segments that contribute to the CNN's diagnoses.

Emotion is a property by which human beings and machines can be differentiated as machines are emotionless while human beings are not. If the emotion of a speaker is recognized then others can interact accordingly. This paper presents a new approach for recognizing all the six basic emotions (Happy, anger, fear, sadness, boredom and neutral) from the speech signals more effectively. To recognize the emotion of a speaker, pitch value and two wavelet packet feature vectors derived from speech signals are used. Principal Component Analysis (PCA) has been applied to reduce the dimension of feature vectors. Random Forest (RF) and Support Vector Machine (SVM) classifiers are trained separately based on these reduced feature vectors. The experimental results show that the accuracy of emotion recognition with Random Forest classifier is 86.11% while with SVM classifier it is 84.41%. Experimentally, it is also found that clean speech of 1 sec duration is sufficient enough to recognize emotion of the speaker.

One of the main challenges that the industry faces when dealing with massive data for failure diagnosis is high dimensionality of such data. This can be tackled by dimensionality reduction method such as principal components analysis, which usually results in an improved fault diagnosis. Other available techniques include auto-encoders and its variants denoising auto-encoders and sparse auto-encoders. Most recently, variational auto-encoders are one of the most promising techniques for unsupervised learning with successful applications in image processing and speech recognition. Differently from other auto-encoder methods, variational auto-encoders use variational inference to generate a latent representation of the data and impose a distribution over the latent variables and the data itself. In this article, we propose a fully unsupervised deep variational auto-encoder-based approach for dimensionality reduction in fault diagnosis and explore the variational auto-encoder capabilities for such a task. This is achieved by comparing the latent representations provided by variational auto-encoders to the ones from principal components analysis as well as when no reduction is performed in ball bearings’ fault classification using vibration signals. To tackle massive sensor data, we propose different architectures for the variational auto-encoder’s encoder and decoder that are based on deep neural networks and deep convolutional neural networks. Experiments are also carried out by varying the data preprocessing methods for generating spectrograms and hand-engineering features as well as the use of raw vibration data, the architecture of the neural networks fault classifiers operating on the latent representations from variational auto-encoder and principal components analysis, the degree of data dimension reduction, and the size of the available labeled data used for training the fault classifiers. The results show that variational auto-encoders are a competent and promising tool for dimensionality reduction for use in fault diagnosis and worth further exploring their capabilities beyond vibration signals of ball bearing elements.

This paper carries on research surrounding the influences produced by dimensionality reduction on machine learning classification effect. Firstly, paper constructs the analysis architecture of data dimension reduction classification, combines the two different unsupervised dimension reduction methods, locally linear embedding (LLE) and principal component analysis (PCA) with the five machine learning classification methods: Gradient Boosting Decision Tree (GBDT), Random Forest, Support Vector Machine (SVM), K-Nearest Neighbor (KNN) and Logistic Regression. And then uses the handwritten digital identification dataset to analyze the classification performance of these five classification methods on different dimension datasets by different dimensionality reduction methods. The analysis shows that using the appropriate dimensionality reduction method for dimensionality reduction classification can effectively improve the classification accuracy; the dimensionality reduction classification effect of non-linear dimensionality reduction method is generally better than the linear dimensionality reduction method; different machine learning classification algorithms have significant differences in the sensitivity of dimensions.

Atrial fibrillation (AF) is a very common arrhythmia that often causes the serious complication of a stroke. The aim was to evaluate the utility of pathological chemoreflexsensitivity (PCHRS) and atrial late potentials (ALP) to predict AF in follow-up. We investigated a prospective study on the basis of our observation about a PCHRS and ALP in paroxysmal AF. The PCHRS was predefined as a chemoreflexsensitivity below 3.0 ms/mmHg and ALP were predefined as a filtered P-wave duration > or =120 ms and a root mean square voltage of the last 20 ms of the P-wave < or =3.5 microV. A P-wave triggered P-wave signal averaged electrocardiograph (ECG) and chemoreflexsensitivity was performed on 250 consecutive patients who were divided into four groups. Group I consisted of patients with ALP and PCHRS, patients of group II had only ALP, a PCHRS was only present in group III, and patients of group IV had neither ALP nor PCHRS. During the mean follow-up of 37.8 months AF was observed in 10 patients (4%). The patients of the four groups were similar according to clinical baseline characteristics. The incidence of AF was higher in group I (18% of patients) than in group II (6% of patients, P = 0.229) and significantly higher than in group III (3% of patients, P = 0.034) or group IV (1% of patients, P < 0.0001). Patients with ALP and PCHRS showed a 33-fold risk (P < 0.001) for the onset AF. The results of our study suggest that the probability of AF could be predicted with a P-wave signal averaged ECG and an analysis of chemoreflexsensitivity. The predictive power of the combination of ALP and PCHRS seemed not high enough for risk stratification.

Hyperspectral imaging (HSI) is increasingly being used in a wide range of practical applications. High correlations between and within classes, the curse of dimensionality, and overfitting have all piqued the interest of researchers. Because of the excellent feature extraction and good overall performance, deep learning-based algorithms with dimensionality reduction methods have been recommended as a viable alternative for hyperspectral image analysis. Principal component analysis (PCA) and Sparse PCA (SPCA) are commonly used in dimensionality reduction. Because each principal component is a linear mixture of the original variables, principal component analysis (PCA) is difficult to interpret, and SPCA cannot choose the most relevant features. To overcome the issue, a mutual information-based greedy feature selection approach called minimum redundancy maximum relevancy (mRMR) is combined with SPCA (SPCA-mRMR). Many deep learning combination techniques rely on the 2D and 3D Convolutional Neural Network (CNN) methodologies. A fused 3D-2D CNN model is less difficult than a 3D-CNN alone, yet it can also operate effectively in noisy settings with few training examples. Therefore, SPCA-mRMR and fused 3D-2D CNN are used to create a hybrid model called HMC-NET for HSI classification. Several optimization strategies are used to address overfitting and improve classification accuracy. To assess the efficacy of this proposed technique, performance results were compared to highly regarded models with fewer variables.

&amp;lt;p&amp;gt;&amp;lt;strong&amp;gt;Potential of Sentinel-1 and Sentinel-2 data for Mapping Irrigated areas at plot scale&amp;lt;/strong&amp;gt;&amp;lt;/p&amp;gt;&amp;lt;p&amp;gt;Hassan Bazzi &amp;lt;sup&amp;gt;1&amp;lt;/sup&amp;gt;, Nicolas Baghdadi &amp;lt;sup&amp;gt;1&amp;lt;/sup&amp;gt;, Dino Ienco &amp;lt;sup&amp;gt;1&amp;lt;/sup&amp;gt;, Mehrez Zribi &amp;lt;sup&amp;gt;2&amp;lt;/sup&amp;gt;, Hatem Belhouchette &amp;lt;sup&amp;gt;3&amp;lt;/sup&amp;gt;&amp;lt;/p&amp;gt;&amp;lt;p&amp;gt;Irrigation plays a significant role in agricultural production in order to meet the global food requirement under changing climatic conditions. To fulfill the high demand for food with an ever-increasing global population, better planning of irrigation is required. Therefore, more focus is being set on the assessment of irrigation performance for improving water management in order to achieve higher water productivity and increase agricultural water sustainability.&amp;lt;/p&amp;gt;&amp;lt;p&amp;gt;In the context of mapping irrigated areas, we propose an innovative approach to map irrigated areas using Sentinel-1 (S1) SAR (Synthetic Aperture Radar) and Sentinel-2 (S2) optical time series. Our proposed approach is based on the use of S1 and/or S2 time series combined with statistical and mathematical functions such as principal component analysis (PCA) and wavelet transformation (WT). The proposed approach was tested over the Catalonia region, Spain with a dataset containing 126 000 irrigated and 67 000 non-irrigated plots. The novelty of our study resides in eliminating the ambiguity between irrigation and rainfall by comparing between the SAR backscattering signal of each plot and that of the corresponding grid (10 km &amp;amp;#215; 10 km). The potential of S2 images to classify irrigated areas by means of NDVI time series was also investigated in this study. Random forest (RF) and convolutional neural network (CNN) approaches were used to build up classification models using the PCA or WT parameters in three different scenarios: The first using only S1 data, the second using only S2 data, and the third using both S1 and S2 data.&amp;lt;/p&amp;gt;&amp;lt;p&amp;gt;The RF classifiers built using the PCA or WT on the S1 time series perform well in mapping irrigated areas with an accuracy of 90.7% and 89.1% respectively. However, the CNN classification on the S1 temporal series produces a significant overall accuracy of 94.1%. The overall accuracy obtained using the NDVI time series in RF classifier reached 89.5% while that in the CNN reached 91.6%. Finally, the combined use of the SAR and optical data enhanced the accuracy of the RF classification but did not significantly change the overall accuracy of the CNN model.&amp;lt;/p&amp;gt;

An accurate electrocardiogram (ECG) beat classification can benefit the diagnosis of the cardiovascular disease. Deep convolutional neural networks (CNN) can automatically extract valid features from data, which is an effective way for the classification of the ECG beats. However, the fully-connected layer in CNNs requires a fixed input dimension, which limits the CNNs to receive fixed-scale inputs. Signals of different scales are generally processed into the same size by segmentation and downsampling. If information loss occurs during a uniformly-sized process, the classification accuracy will ultimately be affected. To solve this problem, this paper constructs a new CNN framework spatial pyramid pooling (SPP) method, which solves the deficiency caused by the size of input data. The Massachusetts Institute of Technology-Biotechnology (MIT-BIH) arrhythmia database is employed as the training and testing data for the classification of heartbeat signals into six categories. Compared with the traditional method, which may lose a large amount of important information and easy to be over-fitted, the robustness of the proposed method can be guaranteed by extracting data features from different sizes. Experimental results show that the proposed architecture network can extract more high-quality features and exhibits higher classification accuracy (94%) than the traditional deep CNNs (90.4%).

Shell vibration signals generated during grinding have useful information related to ball mill load, while usually contaminated by noises. It is a challenge to recognize load parameters with these signals. In this paper, a novel approach is proposed based on the improved empirical wavelet transform (EWT), refined composite multi-scale dispersion entropy (RCMDE) and fireworks algorithm (FWA) optimized SVM. Firstly, vibration signals are denoised by improved EWT, which uses cubic spline interpolation to calculate envelope spectrum for segmentation. Then, RCMDEs of the denoised signals are calculated as feature vectors. The vectors’ dimensionalities are reduced by principal component analysis (PCA). Finally, a mill load prediction model is established based on the FWA optimized SVM. The reduced feature vectors are fed to the model, thus material-to-ball ratio and filling rate being outputs. Grinding experiments show that the extracted features by RCMDE can effectively distinguish three load states. Meanwhile, experiments also show that FWA reduces the forecasting errors of material-to-balls ratio and filling rate by 1.9% and 2.9% compared with genetic algorithm (GA), as well as by 1.92% and 4.21% compared with particle swarm optimization (PSO) algorithm. It demonstrates that the proposed approach for ball mill load forecasting has high accuracy and stability.

Research on feature vector construction and classification recognition of typical mine microseismic waveforms

Real driving environment EEG-based detection of driving fatigue using the wavelet scattering network

Hyperspectral imaging (HSI) captures detailed spectral and spatial information, making it a versatile tool across various domains. This paper presents a Convolutional Neural Network (CNN)based method for vein detection from hyperspectral images of human hands. The proposed method applies a dimensionality reduction technique to the input HSI to extract its features and reduce its dimensionality. A CNN model is then trained and used to identify the location of vein pixels in the input image. Three dimensionality reduction methods, namely Principal Component Analysis (PCA), Folded-PCA (FPCA), and Inter-band Correlation and Clustering using the K-means clustering method (ICC_k-means), were used to reduce the dimensionality of the input image. A 3D-CNN model is then trained to detect the location of veins’ pixels in the input HSI. A 3D-CNN model was trained on dimensionality reduced HSI data to accurately identify the location of veins’ pixels in the input HSI. Experimental results were generated using the HyperVein dataset. The dataset images were randomly divided into training, validation, and test sets. Experimental results show that the proposed method using the ICC k-means dimensionality reduction technique achieves the highest accuracy, precision, recall, false positive rate, false negative rate, and receiver operating characteristics compared to when PCA and FPCA methods are used.

Background Risk stratification of heart failure (HF) in individuals without prevalent heart disease is crucial for early intervention and prevention. Electrocardiogram (ECG)-based artificial intelligence models have shown promise when predicting HF events at long follow-up periods. However, they are still limited by imbalanced case-control data and small labelled datasets. Moreover, the electrophysiological mechanisms underlying HF remain understudied. Purpose We compared the performance of a baseline convolutional neural network (CNN) model designed to predict HF events from the ECG at 8 different timepoints (1, 2, 3, 4, 6, 8, 10 &amp; 12 years after the ECG acquisition), with a fine-tuned self-supervised model, pre-trained using a large amount of unlabelled ECGs. Subsequently, we investigated the genetic architecture of the prediction from the strongest model. Methods The baseline model consisted of 6 convolutional layers, an attentive pooling layer and 3 fully connected layers. The fine-tuned self-supervised model was composed of a frozen self-supervised feature extraction block, and a trainable classification block. The self-supervised block, composed of 6 convolutional layers, aimed to extract relevant features from the ECG and was trained with 637,438 publicly available ECGs. The classification block, composed of 1 convolutional layer, an attentive pooling layer and 3 fully connected layers, used a loss function designed to optimize the prediction of HF at the 8 timepoints. The models were trained and tested using 86,567 ECGs from UK Biobank (69,257 for training and 17,310 for test). Then, we performed a genome-wide association study (GWAS) on the standardized output at 8 years of the model with the highest area under the curve (AUC), with sex, age, body mass index, genetic array and the RR interval. Results The fine-tuned self-supervised model showed a better performance across all follow-up times when compared to the baseline model, increasing the AUC 7% on average (Figure 1). The GWAS revealed 27 candidate genes at 23 genome-wide significant loci (P &amp;lt; 5 x 10-8), 19 previously proposed for ECG traits, 9 of these for cardiac structural or functional traits (6 for HF), and 4 new (PATJ, BOLL, NATA/PTGIS and CCDC63). We found these genes were significantly enriched in cardiac conduction and regulation of cardiac contraction processes (Figure 2). Conclusions The use of features derived from a pretrained self-supervised model improved the performance of the CNN model, allowing a better prediction of HF at long follow-up periods. Also, the genetics underlying the ECG-based prediction of HF indicated candidate genes previously reported for ECG and arrhythmia phenotypes, highlighting electrophysiological mechanisms implicated in HF. Our findings support the integration of deep learning with genetic analyses to develop more accurate and clinically valuable risk prediction tools for early intervention and prevention of HF.AUC of the two CNN models Significantly enriched mechanisms