RESUMO
In event-related potential (ERP) analysis, it is commonly assumed that individual trials from a subject share similar properties and originate from comparable neural sources, allowing reliable interpretation of group-averages. Nevertheless, traditional group-level ERP analysis methods, including cluster analysis, often overlook critical information about individual subjects' neural processes due to using fixed measurement intervals derived from averaging. We developed a multi-set consensus clustering pipeline to examine cognitive processes at the individual subject level. Initially, consensus clustering from diverse methods was applied to single-trial EEG epochs of individual subjects. Subsequently, a second level of consensus clustering was performed across the trials of each subject. A newly modified time window determination method was then employed to identify individual subjects' ERP(s) of interest. We validated our method with simulated data for ERP components N2 and P3, and real data from a visual oddball task to confirm the P3 component. Our findings revealed that estimated time windows for individual subjects provide precise ERP identification compared to fixed time windows across all subjects. Additionally, Monte Carlo simulations with synthetic single-trial data demonstrated stable scores for the N2 and P3 components, confirming the reliability of our method. The proposed method enhances the examination of brain-evoked responses at the individual subject level by considering single-trial EEG data, thereby extracting mutual information relevant to the neural process. This approach offers a significant improvement over conventional ERP analysis, which relies on the averaging mechanism and fixed measurement interval.
RESUMO
Noninvasive inspection of layered structures has remained a long-standing challenge for time-resolved imaging techniques, where both resolution and contrast are compromised by prominent signal attenuation, interlayer reflections, and dispersion. Our method based on terahertz (THz) time-domain spectroscopy overcomes these limitations by offering fine resolution and a broadband spectrum to efficiently extract hidden structural and content information from layered structures. We exploit local symmetrical characteristics of reflected THz pulses to determine the location of each layer, and apply a statistical process in the spatiotemporal domain to enhance the image contrast. Its superior performance is evidenced by the extraction of alphabetic characters in 26-layer subwavelength papers as well as layer reconstruction and debonding inspection in the conservation of Terra-Cotta Warriors. Our method enables accurate structure reconstruction and high-contrast imaging of layered structures at ultralow signal-to-noise ratio, which holds great potential for internal inspection of cultural artifacts, electronic components, coatings, and composites with dozens of submillimeter layers.
RESUMO
Phages are the functional viruses that infect bacteria and they play important roles in microbial communities and ecosystems. Phage research has attracted great attention due to the wide applications of phage therapy in treating bacterial infection in recent years. Metagenomics sequencing technique can sequence microbial communities directly from an environmental sample. Identifying phage sequences from metagenomic data is a vital step in the downstream of phage analysis. However, the existing methods for phage identification suffer from some limitations in the utilization of the phage feature for prediction, and therefore their prediction performance still need to be improved further. In this article, we propose a novel deep neural network (called MetaPhaPred) for identifying phages from metagenomic data. In MetaPhaPred, we first use a word embedding technique to encode the metagenomic sequences into word vectors, extracting the latent feature vectors of DNA words. Then, we design a deep neural network with a convolutional neural network (CNN) to capture the feature maps in sequences, and with a bi-directional long short-term memory network (Bi-LSTM) to capture the long-term dependencies between features from both forward and backward directions. The feature map consists of a set of feature patterns, each of which is the weighted feature extracted by a convolution filter with convolution kernels in the CNN slide along the input feature vectors. Next, an attention mechanism is used to enhance contributions of important features. Experimental results on both simulated and real metagenomic data with different lengths demonstrate the superiority of the proposed MetaPhaPred over the state-of-the-art methods in identifying phage sequences.
Assuntos
Bacteriófagos , Microbiota , Bacteriófagos/genética , Redes Neurais de Computação , Algoritmos , Metagenoma/genéticaRESUMO
Convolutional neural networks (CNNs) have achieved significant success in medical image segmentation. However, they also suffer from the requirement of a large number of parameters, leading to a difficulty of deploying CNNs to low-source hardwares, e.g., embedded systems and mobile devices. Although some compacted or small memory-hungry models have been reported, most of them may cause degradation in segmentation accuracy. To address this issue, we propose a shape-guided ultralight network (SGU-Net) with extremely low computational costs. The proposed SGU-Net includes two main contributions: it first presents an ultralight convolution that is able to implement double separable convolutions simultaneously, i.e., asymmetric convolution and depthwise separable convolution. The proposed ultralight convolution not only effectively reduces the number of parameters but also enhances the robustness of SGU-Net. Secondly, our SGUNet employs an additional adversarial shape-constraint to let the network learn shape representation of targets, which can significantly improve the segmentation accuracy for abdomen medical images using self-supervision. The SGU-Net is extensively tested on four public benchmark datasets, LiTS, CHAOS, NIH-TCIA and 3Dircbdb. Experimental results show that SGU-Net achieves higher segmentation accuracy using lower memory costs, and outperforms state-of-the-art networks. Moreover, we apply our ultralight convolution into a 3D volume segmentation network, which obtains a comparable performance with fewer parameters and memory usage. The available code of SGUNet is released at https://github.com/SUST-reynole/SGUNet.
RESUMO
Popular semi-supervised medical image segmentation networks often suffer from error supervision from unlabeled data since they usually use consistency learning under different data perturbations to regularize model training. These networks ignore the relationship between labeled and unlabeled data, and only compute single pixel-level consistency leading to uncertain prediction results. Besides, these networks often require a large number of parameters since their backbone networks are designed depending on supervised image segmentation tasks. Moreover, these networks often face a high over-fitting risk since a small number of training samples are popular for semi-supervised image segmentation. To address the above problems, in this paper, we propose a novel adversarial self-ensembling network using dynamic convolution (ASE-Net) for semi-supervised medical image segmentation. First, we use an adversarial consistency training strategy (ACTS) that employs two discriminators based on consistency learning to obtain prior relationships between labeled and unlabeled data. The ACTS can simultaneously compute pixel-level and image-level consistency of unlabeled data under different data perturbations to improve the prediction quality of labels. Second, we design a dynamic convolution-based bidirectional attention component (DyBAC) that can be embedded in any segmentation network, aiming at adaptively adjusting the weights of ASE-Net based on the structural information of input samples. This component effectively improves the feature representation ability of ASE-Net and reduces the overfitting risk of the network. The proposed ASE-Net has been extensively tested on three publicly available datasets, and experiments indicate that ASE-Net is superior to state-of-the-art networks, and reduces computational costs and memory overhead. The code is available at: https://github.com/SUST-reynole/ASE-Nethttps://github.com/SUST-reynole/ASE-Net.
Assuntos
Aprendizado de Máquina Supervisionado , Processamento de Imagem Assistida por Computador , IncertezaRESUMO
Averaging amplitudes over consecutive time samples (i.e., time window) is widely used to calculate the peak amplitude of event-related potentials (ERPs). Cluster analysis of the spatio-temporal ERP data is a promising tool to determine the time window of an ERP of interest. However, determining an appropriate number of clusters to optimally represent ERPs is still challenging. Here, we develop a new method to estimate the optimal number of clusters utilizing consensus clustering. Various polarity dependent clustering methods, namely, k-means, hierarchical clustering, fuzzy c-means, self-organizing map, spectral clustering, and Gaussian mixture model, are used to configure consensus clustering after assessing them individually. When a range of clusters is applied many times, the optimal number of clusters should correspond to the expectation, which is the average of the obtained mean inner-similarities of estimated time windows across all conditions and groups converge in the satisfactory thresholds. In order to assess our method, the proposed method has been applied to simulated data and prospective memory experiment ERP data aimed to qualify N2 and P3, and N300 and prospective positivity components, respectively. The results of determining the optimal number of clusters meet at six cluster maps for both ERP data. In addition, our results revealed that the proposed method could be reliably applied to ERP data to determine the appropriate time window for the ERP of interest when the measurement interval is not accurately defined.
Assuntos
Potenciais Evocados , Memória Episódica , Humanos , Análise por Conglomerados , Algoritmos , Análise Espaço-Temporal , Eletroencefalografia/métodosRESUMO
As failures of rolling bearings lead to major failures in rotating machines, recent vibration-based rolling bearing fault diagnosis techniques are focused on obtaining useful fault features from the huge collection of raw data. However, too many features reduce the classification accuracy and increase the computation time. This paper proposes an effective feature selection technique based on intrinsic dimension estimation of compressively sampled vibration signals. First, compressive sampling (CS) is used to get compressed measurements from the collected raw vibration signals. Then, a global dimension estimator, the geodesic minimal spanning tree (GMST), is employed to compute the minimal number of features needed to represent efficiently the compressively sampled signals. Finally, a feature selection process, combining the stochastic proximity embedding (SPE) and the neighbourhood component analysis (NCA), is used to select fewer features for bearing fault diagnosis. With regression analysis-based predictive modelling technique and the multinomial logistic regression (MLR) classifier, the selected features are assessed in two case studies of rolling bearings vibration signals under different working loads. The experimental results demonstrate that the proposed method can successfully select fewer features, with which the MLR-based trained model achieves high classification accuracy and significantly reduced computation times compared to published research.
RESUMO
With the advancement in machine learning, researchers continue to devise and implement effective intelligent methods for fraud detection in the financial sector. Indeed, credit card fraud leads to billions of dollars in losses for merchants every year. In this paper, a multi-classifier framework is designed to address the challenges of credit card fraud detections. An ensemble model with multiple machine learning classification algorithms is designed, in which the Behavior-Knowledge Space (BKS) is leveraged to combine the predictions from multiple classifiers. To ascertain the effectiveness of the developed ensemble model, publicly available data sets as well as real financial records are employed for performance evaluations. Through statistical tests, the results positively indicate the effectiveness of the developed model as compared with the commonly used majority voting method for combination of predictions from multiple classifiers in tackling noisy data classification as well as credit card fraud detection problems.
Assuntos
Aprendizado de MáquinaRESUMO
Open circuit failure mode in insulated-gate bipolar transistors (IGBT) is one of the most common faults in modular multilevel converters (MMCs). Several techniques for MMC fault diagnosis based on threshold parameters have been proposed, but very few studies have considered artificial intelligence (AI) techniques. Using thresholds has the difficulty of selecting suitable threshold values for different operating conditions. In addition, very little attention has been paid to the importance of developing fast and accurate techniques for the real-life application of open-circuit failures of IGBT fault diagnosis. To achieve high classification accuracy and reduced computation time, a fault diagnosis framework with a combination of the AC-side three-phase current, and the upper and lower bridges' currents of the MMCs to automatically classify health conditions of MMCs is proposed. In this framework, the principal component analysis (PCA) is used for feature extraction. Then, two classification algorithms-multiclass support vector machine (SVM) based on error-correcting output codes (ECOC) and multinomial logistic regression (MLR)-are used for classification. The effectiveness of the proposed framework is validated by a two-terminal simulation model of the MMC-high-voltage direct current (HVDC) transmission power system using PSCAD/EMTDC software. The simulation results demonstrate that the proposed framework is highly effective in diagnosing the health conditions of MMCs compared to recently published results.
Assuntos
Inteligência Artificial , Máquina de Vetores de Suporte , Algoritmos , Simulação por Computador , Análise de Componente PrincipalRESUMO
Stroke has become a leading cause of death and long-term disability in the world with no effective treatment. Deep learning-based approaches have the potential to outperform existing stroke risk prediction models, but they rely on large well-labeled data. Due to the strict privacy protection policy in health-care systems, stroke data is usually distributed among different hospitals in small pieces. In addition, the positive and negative instances of such data are extremely imbalanced. Transfer learning can solve small data issue by exploiting the knowledge of a correlated domain, especially when multiple source of data are available. In this work, we propose a novel Hybrid Deep Transfer Learning-based Stroke Risk Prediction (HDTL-SRP) scheme to exploit the knowledge structure from multiple correlated sources (i.e., external stroke data, chronic diseases data, such as hypertension and diabetes). The proposed framework has been extensively tested in synthetic and real-world scenarios, and it outperforms the state-of-the-art stroke risk prediction models. It also shows the potential of real-world deployment among multiple hospitals aided with 5 G/B5G infrastructures.
Assuntos
Diabetes Mellitus , Acidente Vascular Cerebral , Hospitais , Humanos , Aprendizado de MáquinaRESUMO
Fault detection and classification are two of the challenging tasks in Modular Multilevel Converters in High Voltage Direct Current (MMC-HVDC) systems. To directly classify the raw sensor data without certain feature extraction and classifier design, a long short-term memory (LSTM) neural network is proposed and used for seven states of the MMC-HVDC transmission power system simulated by Power Systems Computer Aided Design/Electromagnetic Transients including DC (PSCAD/EMTDC). It is observed that the LSTM method can detect faults with 100% accuracy and classify different faults as well as provide promising fault classification performance. Compared with a bidirectional LSTM (BiLSTM), the LSTM can get similar classification accuracy, requiring less training time and testing time. Compared with Convolutional Neural Networks (CNN) and AutoEncoder-based deep neural networks (AE-based DNN), the LSTM method can get better classification accuracy around the middle of the testing data proportion, but it needs more training time.
Assuntos
Memória de Curto Prazo , Redes Neurais de Computação , Eletricidade , Memória de Longo PrazoRESUMO
In view of the fact that vibration signals of rolling bearings are much contaminated by noise in the early failure period, this paper presents a new denoising SVD-VMD method by combining singular value decomposition (SVD) and variational mode decomposition (VMD). SVD is used to determine the structure of the underlying model, which is referred to as signal and noise subspaces, and VMD is used to decompose the original signal into several band-limited modes. Then the effective components are selected from these modes to reconstruct the denoised signal according to the difference spectrum (DS) of singular values and kurtosis values. Simulated signals and experimental signals of roller bearing faults have been analyzed using this proposed method and compared with SVD-DS. The results demonstrate that the proposed method can effectively retain the useful signals and denoise the bearing signals in extremely noisy backgrounds.
Assuntos
Algoritmos , Vibração , Ruído , Modalidades de Fisioterapia , Razão Sinal-RuídoRESUMO
Discovering DNA-protein binding sites, also known as motif discovery, is the foundation for further analysis of transcription factors (TFs). Deep learning algorithms such as convolutional neural networks (CNN) have been introduced to motif discovery task and have achieved state-of-art performance. However, due to the limitations of CNN, motif discovery methods based on CNN do not take full advantage of large-scale sequencing data generated by high-throughput sequencing technology. Hence, in this paper we propose multi-scale capsule network architecture (MSC) integrating multi-scale CNN, a variant of CNN able to extract motif features of different lengths, and capsule network, a novel type of artificial neural network architecture aimed at improving CNN. The proposed method is tested on real ChIP-seq datasets and the experimental results show a considerable improvement compared with two well-tested deep learning-based sequence model, DeepBind and Deepsea.
Assuntos
Sítios de Ligação/genética , Biologia Computacional/métodos , Proteínas de Ligação a DNA , Aprendizado Profundo , Fatores de Transcrição , Algoritmos , Sequenciamento de Cromatina por Imunoprecipitação , Proteínas de Ligação a DNA/química , Proteínas de Ligação a DNA/genética , Proteínas de Ligação a DNA/metabolismo , Ligação Proteica/genética , Análise de Sequência de Proteína , Fatores de Transcrição/química , Fatores de Transcrição/genética , Fatores de Transcrição/metabolismoRESUMO
Clustering is a promising tool for grouping the sequence of similar time-points aimed to identify the attention blocks in spatiotemporal event-related potentials (ERPs) analysis. It is most likely to elicit the appropriate time window for ERP of interest if a suitable clustering method is applied to spatiotemporal ERP. However, how to reliably estimate a proper time window from entire individual subjects' data is still challenging. In this study, we developed a novel multiset consensus clustering method in which several clustering results of multiple subjects were combined to retrieve the best fitted clustering for all the subjects within a group. Then, the obtained clustering was processed by a newly proposed time-window detection method to determine the most suitable time window for identifying the ERP of interest in each condition/group. Applying the proposed method to the simulated ERP data and real data indicated that the brain responses from the individual subjects can be collected to determine a reliable time window for different conditions/groups. Our results revealed more precise time windows to identify N2 and P3 components in the simulated data compared to the state-of-the-art methods. Additionally, our proposed method achieved more robust performance and outperformed statistical analysis results in the real data for N300 and prospective positivity components. To conclude, the proposed method successfully estimates the time window for ERP of interest by processing the individual data, offering new venues for spatiotemporal ERP processing.
RESUMO
In this paper, we explore learning methods to improve the performance of the open-circuit fault diagnosis of modular multilevel converters (MMCs). Two deep learning methods, namely, convolutional neural networks (CNN) and auto encoder based deep neural networks (AE-based DNN), as well as stand-alone SoftMax classifier are explored for the detection and classification of faults of MMC-based high voltage direct current converter (MMC-HVDC). Only AC-side three-phase current and the upper and lower bridges' currents of the MMCs are used directly in our proposed approaches without any explicit feature extraction or feature subset selection. The two-terminal MMC-HVDC system is implemented in Power Systems Computer-Aided Design/Electromagnetic Transients including DC (PSCAD/EMTDC) to verify and compare our methods. The simulation results indicate CNN, AE-based DNN, and SoftMax classifier can detect and classify faults with high detection accuracy and classification accuracy. Compared with CNN and AE-based DNN, the SoftMax classifier performed better in detection and classification accuracy as well as testing speed. The detection accuracy of AE-based DNN is a little better than CNN, while CNN needs less training time than the AE-based DNN and SoftMax classifier.
RESUMO
BACKGROUND: Red blood cells are essential for modern medicine but managing their collection and supply to cope with fluctuating demands represents a major challenge. As deterministic models based on predicted population changes have been problematic, there remains a need for more precise and reliable prediction of use. Here, we develop three new time-series methods to predict red cell use 4 to 52 weeks ahead. STUDY DESIGN AND METHODS: From daily aggregates of red blood cell (RBC) units issued from 2005 to 2011 from the NHS Blood and Transplant, we generated a new set of non-overlapping weekly data by summing the daily data over 7 days and derived the average blood use per week over 4-week and 52-week periods. We used three new methods for linear prediction of blood use by computing the coefficients using Minimum Mean Squared Error (MMSE) algorithm. RESULTS: We optimized the time-window size, order of the prediction, and order of the polynomial fit for our data set. By exploiting the annual periodicity of the data, we achieved significant improvements in long-term predictions, as well as modest improvements in short-term predictions. The new methods predicted mean RBC use with a standard deviation of the percentage error of 2.5% for 4 weeks ahead and 3.4% for 52 weeks ahead. CONCLUSION: This paradigm allows short- and long-term prediction of RBC use and could provide reliable and precise prediction up to 52 weeks ahead to improve the efficiency of blood services and sufficiency of blood supply with reduced costs.
Assuntos
Algoritmos , Doadores de Sangue/provisão & distribuição , Segurança do Sangue , Transfusão de Sangue/tendências , Bases de Dados Factuais , Inglaterra , Feminino , Humanos , Modelos Lineares , MasculinoRESUMO
BACKGROUND: Blood products are essential for modern medicine, but managing their collection and supply in the face of fluctuating demands represents a major challenge. As deterministic models based on predicted changes in population have been problematic, there remains a need for more precise and reliable prediction of demands. Here, we propose a paradigm incorporating four different time-series methods to predict red blood cell (RBC) issues 4 to 24 weeks ahead. STUDY DESIGN AND METHODS: We used daily aggregates of RBC units issued from 2005 to 2011 from the National Health Service Blood and Transplant. We generated a new set of nonoverlapping weekly data by summing the daily data over 7 days and derived the average blood issues per week over 4-week periods. We used four methods for linear prediction of blood demand by computing the coefficients with the minimum mean squared error and weighted least squares error algorithms. RESULTS: We optimized the time-window size, order of the prediction, and order of the polynomial fit for our data set. The four time-series methods, essentially using different weightings to data points, gave very similar results and predicted mean RBC issues with a standard deviation of the percentage error of 3.0% for 4 weeks ahead and 4.0% for 24 weeks ahead. CONCLUSION: This paradigm allows prediction of demand for RBCs and could be developed to provide reliable and precise prediction up to 24 weeks ahead to improve the efficiency of blood services and sufficiency of blood supply with reduced costs.
Assuntos
Transfusão de Eritrócitos/métodos , Eritrócitos , Algoritmos , Inglaterra , Humanos , Análise dos Mínimos QuadradosRESUMO
Two of the data modelling techniques - polynomial representation and time-series representation - are explored in this paper to establish their connections and differences. All theoretical studies are based on uniformly sampled data in the absence of noise. This paper proves that all data from an underlying polynomial model of finite degree [Formula: see text] can be represented perfectly by an autoregressive time-series model of order [Formula: see text] and a constant term [Formula: see text] as in equation (2). Furthermore, all polynomials of degree [Formula: see text] are shown to give rise to the same set of time-series coefficients of specific forms with the only possible difference being in the constant term [Formula: see text]. It is also demonstrated that time-series with either non-integer coefficients or integer coefficients not of the aforementioned specific forms represent polynomials of infinite degree. Six numerical explorations, with both generated data and real data, including the UK data and US data on the current Covid-19 incidence, are presented to support the theoretical findings. It is shown that all polynomials of degree [Formula: see text] can be represented by an all-pole filter with [Formula: see text] repeated roots (or poles) at [Formula: see text]. Theoretically, all noise-free data representable by a finite order all-pole filter, whether they come from finite degree or infinite degree polynomials, can be described exactly by a finite order AR time-series; if the values of polynomial coefficients are not of special interest in any data modelling, one may use time-series representations for data modelling.
RESUMO
With the unprecedented development of the Internet, it also brings the challenge of Internet Addiction (IA), which is hard to diagnose and cure according to the state-of-art research. In this study, we explored the feasibility of machine learning methods to detect IA. We acquired a dataset consisting of 2397 Chinese college students from the University (Age: 19.17⯱â¯0.70, Male: 64.17%) who completed Brief Self Control Scale (BSCS), the 11th version of Barratt Impulsiveness Scale (BIS-11), Chinese Big Five Personality Inventory (CBF-PI) and Chen Internet Addiction Scale (CIAS), where CBF-PI includes five sub-features (Openness, Extraversion, Conscientiousness, Agreeableness, and Neuroticism) and BSCS includes three sub-features (Attention, Motor and Non-planning). We applied Student's t-test on the dataset for feature selection and Support Vector Machines (SVMs) including C-SVM and ν-SVM with grid search for the classification and parameters optimization. This work illustrates that SVM is a reliable method for the assessment of IA and questionnaire data analysis. The best detection performance of IA is 96.32% which was obtained by C-SVM in the 6-feature dataset without normalization. Finally, the BIS-11, BSCS, Motor, Neuroticism, Non-planning, and Conscientiousness are shown to be promising features for the detection of IA.