Search

A robust decoding model that can efficiently deal with the subject and period variation is urgently needed to apply the brain–computer interface (BCI) system. The performance of most electroencephalogram (EEG) decoding models depends on the characteristics of specific subjects and periods, which require calibration and training with annotated data prior to application. However, this situation will become unacceptable as it would be difficult for subjects to collect data for an extended period, especially in the rehabilitation process of disability based on motor imagery (MI). To address this issue, we propose an unsupervised domain adaptation framework called iterative self-training multisubject domain adaptation (ISMDA) that focuses on the offline MI task. First, the feature extractor is purposefully designed to map the EEG to a latent space of discriminative representations. Second, the attention module based on dynamic transfer matches the source domain and target domain samples with a higher coincidence degree in latent space. Then, an independent classifier oriented to the target domain is employed in the first stage of the iterative training process to cluster the samples of the target domain through similarity. Finally, a pseudolabel algorithm based on certainty and confidence is employed in the second stage of the iterative training process to adequately calibrate the error between prediction and empirical probabilities. To evaluate the effectiveness of the model, extensive testing has been performed on three publicly available MI datasets, the BCI IV IIa, the High gamma dataset, and Kwon et al. datasets. The proposed method achieved 69.51%, 82.38%, and 90.98% cross-subject classification accuracy on the three datasets, which outperforms the current state-of-the-art offline algorithms. Meanwhile, all results demonstrated that the proposed method could address the main challenges of the offline MI paradigm.

Electroencephalogram (EEG) has been widely used in driver fatigue recognition due to its high-resolution and noninvasive characteristics in recent years. However, EEG signals have strong variability and significant individual differences, which put forward extremely high requirements on the generalization ability of the model in practical use. To address these issues, a dynamic threshold distribution domain adaptation network (DTDDAN) is proposed to classify the EEG signals for driver fatigue recognition. In the model, the domain discriminator is introduced to alleviate the marginal distribution difference of different domains, and a new loss, Jensen–Shannon loss (JS loss), is applied to relieve the conditional distribution difference of different domains to learn cross-subject invariant deep features of EEG signals. In addition, we propose a dynamic threshold pseudo label strategy to assign pseudo labels to the samples of the target domain in the training process of the model. This strategy can make sure that the high-quality unlabeled data of the target domain contribute to the model training and the model can fully learn the information of the target domain. With extensive experiments on our self-constructed EEG-based fatigue driving data set, competitive performance is achieved for the cross-subject classification task, with average classification sensitivity, specificity, and accuracy of 89.8%, 93.8%, and 92.2%, respectively. And it is shown that our model can effectively extract cross-subject invariant EEG signals deep features by the experimental results.

Epilepsy is a recurrent neurological disorder, and nearly 30% of patients with epilepsy continue to experience symptoms despite taking anti-epileptic drugs. Predicting epileptic seizures enables patients to proactively take preventive measures against potential harm. Higher accuracy (ACC) of seizure prediction would lead to a reduced incidence rate and decreased labor and resource consumption. In this study, we propose a hybrid long short-term memory (LSTM)-Transformer model for predicting epileptic seizures using scalp electroencephalogram (EEG) data. Time–frequency features are extracted through the short-time Fourier transform (STFT) applied to EEG signals, which are then inputted into the model to distinguish the interictal state and the preictal state. Our approach combines the long-distance dependence capability of the Transformer with the advantages of LSTM in processing variable-length information, resulting in more robust and informative feature extraction. We evaluate our proposed method on the Children’s Hospital Boston-MIT (CHB-MIT) dataset and conduct quantitative comparisons with recent methods. The results demonstrate that our method achieves a sensitivity of 99.75%, a false prediction rate (FPR) of 0/h, and an area under the curve (AUC) of 99.39%. This novel approach provides valuable insights for epilepsy prediction.

Characterizing the nonlinear dynamic behaviors of gas–liquid two-phase flow is a challenging but rewarding research topic. In this article, a novel multilayer visibility graph-based ordinal network (MVGON) is proposed for exploring the gas–liquid flow behaviors. In particular, the vertical upward gas–liquid flow experiments are first conducted in a 50 mm inner-diameter pipe, and multivariate measurement data is acquired. Then, MVGON is inferred from experimental signals. MVGON not only can effectively fuse multivariate time series but also enables to further excavate concise key information based on inheriting system features extracted by visibility graph (VG). For each derived projection network of MVGON, graph energy and average clustering coefficient (ACC) are calculated for quantitatively characterizing the topological structure of MVGON. In addition, permutation entropy (PE) and multivariate pseudo Wigner distribution-based time–frequency distribution are calculated for the three typical flow patterns to support our findings. The results indicate that our MVGON framework allows effectively characterizing the nonlinear dynamic behaviors during the evolution of different gas–liquid flow patterns. Meanwhile, it provides a novel approach for characterizing the complex system dynamics based on multivariate time series analysis.

Brain-computer interface (BCI) establishes an interactive platform by translating brain activity patterns into commands of external devices. BCIs, especially motor imagery (MI)-based BCIs, have injected new vitality into the development of rehabilitation medicine and many other fields. In this work, one convolutional neural network, named as Flashlight-Net model, is proposed for multiclass MI classification. Flashlight-Net model adopts modular design, in which channel fusion module and time domain module ensure the directions of feature extraction, while feature pool module reduces the loss of effective information. Given the multi-frequency nature of the brain, we combine three frequency bands and construct an ensemble Flashlight-Net model. During the model training, by means of transfer learning, pretraining and fine-tuning processes are designed to integrate training samples from multiple subjects. The experimental results on publicly available BCI Competition IV-2a dataset show that the proposed model can achieve good results on all nine subjects, with an average classification accuracy of 81.23% for four classes. All these demonstrate that the proposed Flashlight-Net model can effectively decode multi-channel and multiclass MI signals.

Gas–liquid two-phase flow is vital in industrial production. The transient and unstable nature of two-phase flow makes it a challenging task to accurately measure and analyze the flow parameters. In this article, we first design a dynamic experiment of gas–liquid two-phase flow and use the designed four-sector sensor to obtain the dynamic experimental data. Then, we construct the joint recurrence network from the collected data and calculate the average node degree for characterizing the evolution of gas–liquid two-phase flow. Finally, we propose a recurrence-network-based densely connection neural network for flow parameter measurements in two-phase flow, and in the end, the accuracy of flow pattern identification is 95.17%, and the mean absolute percentage error (MAPE) of our model is less than 2.8% and the mean squared error (mse) is less than 0.0023. At the same time, our method also offers a novel approach for industrial soft measurement models.

Brain tumor segmentation is a key step in brain cancer diagnosis. Segmentation of brain tumor sub-regions, including necrotic, enhancing, and edematous regions, can provide more detailed guidance for clinical diagnosis. Weakly supervised brain tumor segmentation methods have received much attention because they do not require time-consuming pixel-level annotations. However, existing weakly supervised methods focus on the segmentation of the entire tumor region while ignoring the challenging task of multi-label segmentation for the tumor sub-regions. In this article, we propose a weakly supervised approach to solve the multi-label brain tumor segmentation problem. To the best of our knowledge, it's the first end-to-end multi-label weakly supervised segmentation model applied to brain tumor segmentation. With well-designed loss functions and a contrastive learning pre-training process, our proposed Transformer-based segmentation method (WS-MTST) has the ability to perform segmentation of brain tumor sub-regions. We conduct comprehensive experiments and demonstrate that our method reaches the state-of-the-art on the popular brain tumor dataset BraTS (from 2018 to 2020).

Schizophrenia (SCZ) is a chronic and serious mental disorder with a high mortality rate. At present, there is a lack of objective, cost-effective and widely disseminated diagnosis tools to address this mental health crisis globally. Clinical electroencephalogram (EEG) is a noninvasive technique to measure brain activity with high temporal resolution, and accumulating evidence demonstrates that clinical EEG is capable of capturing abnormal SCZ neuropathology. Although EEG-based automated diagnostic tools have obtained impressive performance on individual datasets, the transportability of potential EEG biomarkers in cross-site real-world application is still an open question. To address the challenges of small sample sizes and population heterogeneity, we develop an advanced interpretable deep learning model using multimodal clinical EEG features and demographic information as inputs to graph neural networks, and further propose different transfer learning strategies to adapt to different clinical scenarios. Taking the disease discrimination of health control (HC) and SCZ with 1030 participants as a use case, our model is trained on a small clinical dataset (N= 188, Chinese) and enhanced using a large-scale public dataset (N= 508, American) of adult participants. Cross-site validation from an independent dataset of adult participants (N= 157, Chinese) produced stable performance, with AUCs of 0.793–0.852 and accuracies of 0.786–0.858 for different SCZ prevalence, respectively. In addition, cross-site validation from another dataset of adolescent boys (N= 84, Russian) yielded an AUC of 0.702 and an accuracy of 0.690. Moreover, feature visualization further revealed that the ranking of feature importance varied significantly among different datasets, and that EEG theta and alpha band power appeared to be the most significant and translational biomarkers of SCZ pathology. Overall, our promising results demonstrate the feasibility of SCZ discrimination using EEG biomarkers in multiple clinical settings.

Motor imagery (MI)-based brain-computer interface (BCI) provides a promising solution for the limb rehabilitation of stroke patients. However, due to the inadequate cognition of brain state during MI, the further application of BCI in the field of rehabilitation medicine is limited. In this paper, we focus on the MI tasks of four commonly used upper limb rehabilitation actions. And a complex network based topological analysis method is developed to characterize the brain state evolution mechanism over time. In detail, we first carry out the MI experiments and acquire the EEG signals associated with imagining left/right fist clenching and left/right wrist dorsiflexion. Then a multi-frequency multilayer (MFML) network is constructed, which maps the topological features of four MI-related frequency bands. The brain electrodes are defined as network nodes, and each frequency band corresponds to one single layer. Global clustering coefficient (GC) is introduced to conduct quantitative topological analysis. Experimental results show that the topological structures of brain tend to be more energy consuming over time, suggesting that the cognitive demands of subjects gradually increase during MI. At the same time, it is found that continuous MI experiments improve the mind control ability of subjects, and the distinguishability of MI signals also gradually improves. Our research quantitatively explores the evolution of brain states behind MI, which is expected to guide the follow-up application of MI-BCI in clinical rehabilitation medicine.

Nuclei segmentation is an essential step in the automatic analysis of histopathology images. This segmentation task requires much work to manually generate accurate pixel-level annotations for fully supervised training. To overcome such expensive and hard-to-get pixel-level annotations, in this article, we propose a weakly supervised method for nuclei segmentation only using point annotations based on convolutional neural network (CNN). The proposed method effectively combines a weakly supervised segmentation task and an auxiliary colorization task. The dual input with boundaries and color information maximizes the inherent features of the image. Two types of coarse labels generated from point annotations are applied to provide constraint information for the segmentation task. As an auxiliary task, colorization is incorporated to guide the network to extract effective features and improve the segmentation performance of the network. We evaluate our proposed method on two public nuclei segmentation datasets. The experimental results indicate that our method is superior to other state-of-the-art methods.

The applications of myoelectrical interfaces are majorly limited by the efficacy of decoding motion intent in the electromyographic (EMG) signal. Currently, EMG classification methods often rely substantially on handcrafted features or ignore key channel and interfeature information for classification tasks. To address these issues, a multiscale feature extraction network (MSFEnet) based on channel-spatial attention is proposed to decode the EMG signal for the task of gesture recognition classification. Specifically, we fuse the spatiotemporal characteristics of the EMG signal with different scales. Then, we construct the feature channel attention module and the feature-spatial attention module to capture more key channels features and more key spatial features. To evaluate the efficacy of the proposed method, extensive experiments are conducted on two public data sets: 1) NinaPro DB2 and 2) CapgMyo DB-a. An average accuracy of 86.21%, 90.77%, 92.53%, and 98.85% has been achieved in Exercises B, C, and D of NinaPro DB2 and CapgMyo DB-a, respectively. The experimental results demonstrate that MSFEnet is more capable of extracting temporal and spatial fused features. It performs well in generalization and has higher classification accuracy compared with the state-of-the-art methods.

Gas–liquid two-phase flow is pervasive in industrial processes, and accurate measurement of its parameters can effectively improve industrial efficiency. Void fraction, a key parameter in gas–liquid two-phase flow, plays a crucial role in mixture density determination and flow structure analysis. In this article, a novel multitask learning (MTL) method is proposed to predict the gas void fraction. First, we conduct vertical upward gas–liquid two-phase flow experiments to obtain fluid signals relying on the four-sector distributed conductance sensor. Then, we pair the measurement of gas void fraction and its classification and design a co-attention-based cross-stitch network (CACSNet) to process these two closely related tasks simultaneously. In CACSNet, two parallel embedding branches are adopted to extract task-specific features. There are three feature extractors in each branch to capture in-depth representations of temporal and spatial. To learn an optimal combination of these task-specific features and generate shared representations, we propose cross-stitch embedding with attention modules (CEAMs), which is incorporated the channel co-attention module (CCAM) and the temporal co-attention module (TCAM). Finally, we compare our CACSNet with single-task baseline and other competitive methods, which proves that our framework achieves superior performance in gas void fraction prediction. We also conduct detailed model ablation and parameter analysis to illustrate the efficiency of our proposed structure.

Human emotion recognition could greatly contribute to human–computer interaction with promising applications in artificial intelligence. One of the challenges in recognition tasks is learning effective representations with stable performances from electroencephalogram (EEG) signals. In this article, we propose a novel deep-learning framework, named channel-fused dense convolutional network, for EEG-based emotion recognition. First, we use a 1-D convolution layer to receive weighted combinations of contextual features along the temporal dimension from EEG signals. Next, inspired by state-of-the-art object classification techniques, we employ 1-D dense structures to capture electrode correlations along the spatial dimension. The developed algorithm is capable of handling temporal dependencies and electrode correlations with the effective feature extraction from noisy EEG signals. Finally, we perform extensive experiments based on two popular EEG emotion datasets. Results indicate that our framework achieves prominent average accuracies of 90.63% and 92.58% on the SEED and DEAP datasets, respectively, which both receive better performances than most of the compared studies. The novel model provides an interpretable solution with excellent generalization capacity for broader EEG-based classification tasks.

Background: Alzheimer's Disease (AD) is a progressive neurodegenerative disease with insidious onset, which is difficult to be reversed and cured. Therefore, discovering more precise biological information from neuroimaging biomarkers is crucial for accurate and automatic detection of AD. Methods: We innovatively used a Visibility Graph (VG) to construct the time-dependent brain networks as well as functional connectivity network to investigate the underlying dynamics of AD brain based on functional magnetic resonance imaging. There were 32 AD patients and 29 Normal Controls (NCs) from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database. First, the VG method mapped the time series of single brain region into networks. By extracting topological properties of the networks, the most significant features were selected as discriminant features into a supporting vector machine for classification. Furthermore, in order to detect abnormalities of these brain regions in the whole AD brain, functional connectivity among different brain regions was calculated based on the correlation of regional degree sequences. Results: According to the topology abnormalities exploration of local complex networks, we found several abnormal brain regions, including left insular, right posterior cingulate gyrus and other cortical regions. The accuracy of characteristics of the brain regions extracted from local complex networks was 88.52%. Association analysis demonstrated that the left inferior opercular part of frontal gyrus, right middle occipital gyrus, right superior parietal gyrus and right precuneus played a tremendous role in AD. Conclusion: These results would be helpful in revealing the underlying pathological mechanism of the disease.

Driver fatigue evaluation is of great importance for traffic safety and many intricate factors would exacerbate the difficulty. In this paper, based on the spatial–temporal structure of multichannel electroencephalogram (EEG) signals, we develop a novel EEG-based spatial–temporal convolutional neural network (ESTCNN) to detect driver fatigue. First, we introduce the core block to extract temporal dependencies from EEG signals. Then, we employ dense layers to fuse spatial features and realize classification. The developed network could automatically learn valid features from EEG signals, which outperforms the classical two-step machine learning algorithms. Importantly, we carry out fatigue driving experiments to collect EEG signals from eight subjects being alert and fatigue states. Using 2800 samples under within-subject splitting, we compare the effectiveness of ESTCNN with eight competitive methods. The results indicate that ESTCNN fulfills a better classification accuracy of 97.37% than these compared methods. Furthermore, the spatial–temporal structure of this framework advantages in computational efficiency and reference time, which allows further implementations in the brain–computer interface online systems.

This paper presents the characteristics of a double helix capacitance sensor for measurement of the liquid holdup in horizontal oil–water two-phase flow. The finite element method is used to calculate the sensitivity field of the sensor in a pipe with 20mm inner diameter and the effect of sensor geometry on the distribution of sensitivity field is presented. Then, a horizontal oil–water two-phase flow experiment is carried out to measure the response of the double helix capacitance sensor, in which a novel method is proposed to calibrate the liquid holdup based on three pairs of parallel-wire capacitance probes. The performance of the sensor is analyzed in terms of the flow structures detected by mini-conductance array probes.

Characterizing countercurrent flow structures in an inclined oil-water two-phase flow from one-dimensional measurement is of great importance for model building and sensor design. Firstly, we conducted oil-water two-phase flow experiments in an inclined pipe to measure the conductance signals of three typical water-dominated oil-water flow patterns in inclined flow, i.e., dispersion oil-in-water pseudo-slug flow (PS), dispersion oil-in-water countercurrent flow (CT), and transitional flow (TF). In pseudo-slug flow, countercurrent flow and transitional flow, oil is completely dispersed in water. Then we used magnitude and sign decomposition analysis and multifractal analysis to reveal levels of complexity in different flow patterns. We found that the PS and CT flow patterns both exhibited high complexity and obvious multifractal dynamic behavior, but the magnitude scaling exponent and singularity of the CT flow pattern were less than those of the PS flow pattern; and the TF flow pattern exhibited low complexity and almost monofractal behavior, and its magnitude scaling was close to random behavior. Meanwhile, at short time scales, all sign series of two-phase flow patterns exhibited very similar strong positive correlation; at high time scales, the scaling analysis of sign series showed different anti-correlated behavior. Furthermore, with an increase in oil flow rate, the flow structure became regular, which could be reflected by the decrease in the width of spectrum and the difference in dimensions. The results suggested that different oil-water flow patterns exhibited different nonlinear features, and the varying levels of complexity could well characterize the fluid dynamics underlying different oil-water flow patterns.