Attention-guided context asymmetric fusion networks for the liver tumor segmentation of computed tomography images.

Background: Liver tumor segmentation based on medical imaging is playing an increasingly important role in liver tumor research and individualized therapeutic decision-making. However, it remains a challenging in terms of the accuracy of automatic segmentation of liver tumors. Therefore, we aimed to develop a novel deep neural network for improving the results from the automatic segmentation of liver tumors. Methods: This paper proposes the attention-guided context asymmetric fusion network (AGCAF-Net), combining attention guidance and fusion context modules on the basis of a residual neural network for the automatic segmentation of liver tumors. According to the attention-guided context block (AGCB), the feature map is first divided into multiple small blocks, the local correlation between features is calculated, and then the global nonlocal fusion module (GNFM) is used to obtain the global information between pixels. Additionally, the context pyramid module (CPM) and asymmetric semantic fusion module (AFM) are used to obtain multiscale features and resolve the feature mismatch during feature fusion, respectively. Finally, we used the liver tumor segmentation benchmark (LiTS) dataset to verify the efficiency of our designed network. Results: Our results showed that AGCAF-Net with AFM and CPM is effective in improving the accuracy of liver tumor segmentation, with the Dice coefficient increasing from 82.5% to 84.1%. The segmentation results of liver tumors by AGCAF-Net were superior to those of several state-of-the-art U-net methods, with a Dice coefficient of 84.1%, a sensitivity of 91.7%, and an average symmetric surface distance of 3.52. Conclusions: AGCAF-Net can obtain better matched and accurate segmentation in liver tumor segmentation, thus effectively improving the accuracy of liver tumor segmentation.

Author identification of literary works based on text analysis and deep learning.

With the development of science, speech, picture, and other analysis, problems have been gradually better solved, but the study of Chinese text has been a complex problem to overcome. Chinese text analysis requires not only statistics but also semantic comprehension analysis. Different text types need other language style feature modeling to obtain good recognition results. In this study, we use the deep learning method to construct an automatic text feature extraction model and classify it with the author as a classification label. This study presents a literature author recognition model based on deep learning, which is mainly divided into three phases: text preprocessing, feature extraction, and classification. Each part consists of several small modules or steps. First, we input the corpus to Word2Vec to generate the new word vector. Then, the improved text feature extractor based on CNN and Attention extracts the text features and uses them as the input of the CNN convolution layer. After convolution, the text is combined with bits to get Window Feature Sequence. It is the text feature vector. Next, based on LSTM and Softmax classification output, Window Feature Sequence is used as the input of LSTM to obtain two one-dimensional vectors spliced by concatenate layer. Finally, the result is classified through the fully connected layer, Batch Normalization layer, and Softmax. The performance of the proposed model in recognizing authors of Chinese literature was evaluated using two datasets. In the research process, the data we collected included works of different forms, such as prose and fiction. The research results show that the proposed model can effectively identify author identity. The classification accuracy of our proposed algorithm is significantly better than that of the benchmark model.

CRABR-Net: A Contextual Relational Attention-Based Recognition Network for Remote Sensing Scene Objective.

Remote sensing scene objective recognition (RSSOR) plays a serious application value in both military and civilian fields. Convolutional neural networks (CNNs) have greatly enhanced the improvement of intelligent objective recognition technology for remote sensing scenes, but most of the methods using CNN for high-resolution RSSOR either use only the feature map of the last layer or directly fuse the feature maps from various layers in the "summation" way, which not only ignores the favorable relationship information between adjacent layers but also leads to redundancy and loss of feature map, which hinders the improvement of recognition accuracy. In this study, a contextual, relational attention-based recognition network (CRABR-Net) was presented, which extracts different convolutional feature maps from CNN, focuses important feature content by using a simple, parameter-free attention module (SimAM), fuses the adjacent feature maps by using the complementary relationship feature map calculation, improves the feature learning ability by using the enhanced relationship feature map calculation, and finally uses the concatenated feature maps from different layers for RSSOR. Experimental results show that CRABR-Net exploits the relationship between the different CNN layers to improve recognition performance, achieves better results compared to several state-of-the-art algorithms, and the average accuracy on AID, UC-Merced, and RSSCN7 can be up to 96.46%, 99.20%, and 95.43% with generic training ratios.

A Real Time Method Based on Deep Learning for Reconstructing Holographic Acoustic Fields from Phased Transducer Arrays.

Phased transducer arrays (PTA) can control ultrasonic waves to produce a holographic acoustic field. However, obtaining the phase of the corresponding PTA from a given holographic acoustic field is an inverse propagation problem, which is a mathematically unsolvable nonlinear system. Most of the existing methods use iterative methods, which are complex and time-consuming. To better solve this problem, this paper proposed a novel method based on deep learning to reconstruct the holographic sound field from PTA. For the imbalance and randomness of the focal point distribution in the holographic acoustic field, we constructed a novel neural network structure incorporating attention mechanisms to focus on useful focal point information in the holographic sound field. The results showed that the transducer phase distribution obtained from the neural network fully supports the PTA to generate the corresponding holographic sound field, and the simulated holographic sound field can be reconstructed with high efficiency and quality. The method proposed in this paper has the advantage of real-time performance that is difficult to achieve by traditional iterative methods and has the advantage of higher accuracy compared with the novel AcousNet methods.

An adaptive weighted attention-enhanced deep convolutional neural network for classification of MRI images of Parkinson's disease.

BACKGROUND: Parkinson's disease (PD) is the second prevalent neurological diseases with a significant growth rate in incidence. Convolutional neural networks using structural magnetic resonance images (sMRI) are widely used for PD classification. However, the areas of change in the patient's MRI images are small and unfixed. Thus, capturing the features of the areas accurately where the lesions changed became a problem. METHOD: We propose a deep learning framework that combines multi-scale attention guidance and multi-branch feature processing modules to diagnose PD by learning sMRI T2 slice features. In this scheme, firstly, to achieve effective feature transfer and gradient descent, a deep convolutional neural network framework based on dense block is designed. Next, an Adaptive Weighted Attention algorithm is proposed, whose pursers is to extract multi branch and even diverse features. Finally, Dropout layer and SoftMax layer are added to the network structure to obtain good classification results and rich and diverse feature information. The Dropout layer is used to reduce the number of intermediate features to increase the orthogonality between features of each layer. The activation function SoftMax increases the flexibility of the neural network by increasing the degree of fitting to the training set and converting linear to nonlinear. RESULTS: The best performance of the proposed method an accuracy of 92%, a sensitivity of 94%, specificity of 90% and a F1 score of 95% respectively for identifying PD and HC. CONCLUSION: Experiments show that the proposed method can successfully distinguish PD and NC. Good classification results were obtained in PD diagnosis classification task and compared with advanced research methods.

Multiscale Convolutional Neural Network Based on Channel Space Attention for Gearbox Compound Fault Diagnosis.

Gearboxes are one of the most widely used speed and power transfer elements in rotating machinery. Highly accurate compound fault diagnosis of gearboxes is of great significance for the safe and reliable operation of rotating machinery systems. However, traditional compound fault diagnosis methods treat compound faults as an independent fault mode in the diagnosis process and cannot decouple them into multiple single faults. To address this problem, this paper proposes a gearbox compound fault diagnosis method. First, a multiscale convolutional neural network (MSCNN) is used as a feature learning model, which can effectively mine the compound fault information from vibration signals. Then, an improved hybrid attention module, named the channel-space attention module (CSAM), is proposed. It is embedded into the MSCNN to assign weights to multiscale features for enhancing the feature differentiation processing ability of the MSCNN. The new neural network is named CSAM-MSCNN. Finally, a multilabel classifier is used to output single or multiple labels for recognizing single or compound faults. The effectiveness of the method was verified with two gearbox datasets. The results show that the method possesses higher accuracy and stability than other models for gearbox compound fault diagnosis.

The role of the parietal lobe in task-irrelevant suppression during learning.

BACKGROUND: Attention optimizes the selection of visual information, while suppressing irrelevant visual input through cortical mechanisms that are still unclear. We set to investigate these processes using an attention task with an embedded to-be-ignored interfering visual input. OBJECTIVE: We delivered electrical stimulation to attention-related brain areas to modulate these facilitatory/inhibitory attentional mechanisms. We asked whether overtly training on a task while being covertly exposed to visual features from a visually identical but different task tested at baseline might influence post-training performance on the baseline task. METHODS: In Experiment one, at baseline subjects performed an orientation discrimination (OD) task using a pair of gratings presented at individual's psychophysical threshold. We then trained participants over three-day separate sessions on a temporal order judgment task (TOJ), using the exact same gratings but presented with different time offsets. On the last post-training session we re-tested OD. We coupled training with transcranial random noise stimulation (tRNS) over the parietal cortex, the human middle temporal area or sham, in three separate groups. In Experiment two, subjects performed the same OD task at baseline and post-training, while tRNS was delivered at rest during the same sessions and stimulation conditions as in Experiment one. RESULTS: Results showed that tRNS over parietal cortex facilitated learning of the trained TOJ task. Moreover, we found a detrimental effect on the untrained OD task when subjects received parietal tRNS coupled with training (Experiment one), but a benefit on OD when subjects received stimulation while at rest (Experiment two). CONCLUSIONS: These results clearly indicate that task-irrelevant information is actively suppressed during learning, and that this prioritization mechanism of selection likely resides in the parietal cortex.

A hybrid attentional guidance network for tumors segmentation of breast ultrasound images.

PURPOSE: In recent years, breast cancer has become the greatest threat to women. There are many studies dedicated to the precise segmentation of breast tumors, which is indispensable in computer-aided diagnosis. Deep neural networks have achieved accurate segmentation of images. However, convolutional layers are biased to extract local features and tend to lose global and location information as the network deepens, which leads to a decrease in breast tumors segmentation accuracy. For this reason, we propose a hybrid attention-guided network (HAG-Net). We believe that this method will improve the detection rate and segmentation of tumors in breast ultrasound images. METHODS: The method is equipped with multi-scale guidance block (MSG) for guiding the extraction of low-resolution location information. Short multi-head self-attention (S-MHSA) and convolutional block attention module are used to capture global features and long-range dependencies. Finally, the segmentation results are obtained by fusing multi-scale contextual information. RESULTS: We compare with 7 state-of-the-art methods on two publicly available datasets through five random fivefold cross-validations. The highest dice coefficient, Jaccard Index and detect rate ([Formula: see text]%, [Formula: see text]%, [Formula: see text]% and [Formula: see text]%, [Formula: see text]%, [Formula: see text]%, separately) obtained on two publicly available datasets(BUSI and OASUBD), prove the superiority of our method. CONCLUSION: HAG-Net can better utilize multi-resolution features to localize the breast tumors. Demonstrating excellent generalizability and applicability for breast tumors segmentation compare to other state-of-the-art methods.

Personality coherence as a personality dynamics-related concept.

Extant theoretical models of personality coherence/incoherence do not sufficiently address the challenge of explaining personality coherence dynamics and the role of psychological mechanisms, including temperament and attention. To overcome these limitations, the Complex-System Approach to Personality (C-SAP) postulates that personality coherence is a within-person structure that arises from the functional consistency/inconsistency between personality traits/types, underlain by specific attentional and temperament mechanisms that have integrative and regulatory potential. The dominant (reactive, regulative) function of stimulation processing in temperament types is the foundation for assessing personality coherence. This paper presents a revised, fine-grained model of personality coherence-originally arising from the C-SAP-that is enriched by a focus on personality coherence dynamics in relation to behavioral consistency. The methodological principles necessary for studying personality coherence dynamics are outlined in detail. This paper also addresses: (a) research methods for relating personality coherence/incoherence to behavioral consistency/inconsistency, and (b) situational contexts that are important to these personality dynamics. In addition, personality coherence dynamics in relation to the self and character and the impact of the C-SAP assumption that behaviors are more stable than traits/types on the relation between personality coherence and behavioral consistency are discussed.

Multi-Object Detection in Security Screening Scene Based on Convolutional Neural Network.

The technique for target detection based on a convolutional neural network has been widely implemented in the industry. However, the detection accuracy of X-ray images in security screening scenarios still requires improvement. This paper proposes a coupled multi-scale feature extraction and multi-scale attention architecture. We integrate this architecture into the Single Shot MultiBox Detector (SSD) algorithm and find that it can significantly improve the effectiveness of target detection. Firstly, ResNet is used as the backbone network to replace the original VGG network to improve the feature extraction capability of the convolutional neural network for images. Secondly, a multi-scale feature extraction (MSE) structure is designed to enrich the information contained in the multi-stage prediction feature layer. Finally, the multi-scale attention architecture (MSA) is fused onto the prediction feature layer to eliminate the redundant features' interference and extract effective contextual information. In addition, a combination of Adaptive-NMS and Soft-NMS is used to output the final prediction anchor boxes when performing non-maximum suppression. The results of the experiments show that the improved method improves the mean average precision (mAP) value by 7.4% compared to the original approach. New modules make detection much more accurate while keeping the detection speed the same.

A Radio Environment Map Updating Mechanism Based on an Attention Mechanism and Siamese Neural Networks.

A radio environment map (REM) is an effective spectrum management tool. With the increase in the number of mobile devices, the wireless environment changes more and more frequently, bringing new challenges to REM updates. Traditional update methods usually rely on the amount of data collected for updating without paying attention to whether the wireless environment has changed enough. In particular, a waste of computational resources results from the frequently updated REM when the wireless environment does not change much. When the wireless environment changes a lot, the REM is not updated promptly, resulting in a decrease in REM accuracy. To overcome the above problems, this work combines the Siamese neural network and an attention mechanism in computer vision and proposes an update mechanism based on the amount of wireless environmental change starting from image data. The method compares the newly collected crowdsourced data with the constructed REM in terms of similarity. It uses similarity to measure the necessity of the REM to be updated. The algorithm in this paper can achieve a controlled update by setting a similarity threshold with good controllability. In addition, the effectiveness of the algorithm in detecting changes of the wireless environment has been demonstrated by combing simulation data.

Progressive Two-Stage Network for Low-Light Image Enhancement.

At night, visual quality is reduced due to insufficient illumination so that it is difficult to conduct high-level visual tasks effectively. Existing image enhancement methods only focus on brightness improvement, however, improving image quality in low-light environments still remains a challenging task. In order to overcome the limitations of existing enhancement algorithms with insufficient enhancement, a progressive two-stage image enhancement network is proposed in this paper. The low-light image enhancement problem is innovatively divided into two stages. The first stage of the network extracts the multi-scale features of the image through an encoder and decoder structure. The second stage of the network refines the results after enhancement to further improve output brightness. Experimental results and data analysis show that our method can achieve state-of-the-art performance on synthetic and real data sets, with both subjective and objective capability superior to other approaches.

Video-game play and non-symbolic numerical comparison.

Visual display was used by Stäb and Ilg to examine the abilities of video-game players and non-players to determine simple mathematical abilities.

Pupil tracks statistical regularities: behavioral and neural implications.

Pupillary light reflex adjusts the amount of light reaching the retina. Recent work suggests that the brainstem pupillary light reflex pathway is controlled by the environment's internal models derived from higher-order temporal statistics. This finding has implications at the behavioral and neural levels. Pupillary changes in response to statistical regularities could be a metric constituting the precision with which the internal models are represented. These pupillary changes may aid in information processing through attentional mechanisms. One possible region that mediates descending cognitive inputs to pupil cycling is locus coeruleus. Here we propose a unified framework of locus coeruleus' role in modulating pupillary change, which successfully explains current and previous findings. The locus coeruleus could have multiple subsystems selectively (but not exclusively) driven by behavioral relevance and statistical learning to regulate pupillary change.

Emotional contexts modulate intentional memory suppression of neutral faces: Insights from ERPs.

The main goal of present work is to gain new insight into the temporal dynamics underlying the voluntary memory control for neutral faces associated with neutral, positive and negative contexts. A directed forgetting (DF) procedure was used during the recording of EEG to answer the question whether is it possible to forget a face that has been encoded within a particular emotional context. A face-scene phase in which a neutral face was showed in a neutral or emotional scene (positive, negative) was followed by the voluntary memory cue (cue phase) indicating whether the face had to-be remember or to-be-forgotten (TBR and TBF). Memory for faces was then assessed with an old/new recognition task. Behaviorally, we found that it is harder to suppress faces-in-positive-scenes compared to faces-in-negative and neutral-scenes. The temporal information obtained by the ERPs showed: 1) during the face-scene phase, the Late Positive Potential (LPP), which indexes motivated emotional attention, was larger for faces-in-negative-scenes compared to faces-in-neutral-scenes. 2) Remarkably, during the cue phase, ERPs were significantly modulated by the emotional contexts. Faces-in-neutral scenes showed an ERP pattern that has been typically associated to DF effect whereas faces-in-positive-scenes elicited the reverse ERP pattern. Faces-in-negative scenes did not show differences in the DF-related neural activities but larger N1 amplitude for TBF vs. TBR faces may index early attentional deployment. These results support the hypothesis that the pleasantness or unpleasantness of the contexts (through attentional broadening and narrowing mechanisms, respectively) may modulate the effectiveness of intentional memory suppression for neutral information.

Analysis of eye movements in the judgment of enjoyment and non-enjoyment smiles.

Enjoyment smiles are more often associated with the simultaneous presence of the Cheek raiser and Lip corner puller action units, and these units' activation is more often symmetric. Research on the judgment of smiles indicated that individuals are sensitive to these types of indices, but it also suggested that their ability to perceive these specific indices might be limited. The goal of the current study was to examine perceptual-attentional processing of smiles by using eye movement recording in a smile judgment task. Participants were presented with three types of smiles: a symmetric Duchenne, a non-Duchenne, and an asymmetric smile. Results revealed that the Duchenne smiles were judged happier than those with characteristics of non-enjoyment. Asymmetric smiles were also judged happier than the non-Duchenne smiles. Participants were as effective in judging the latter smiles as not really happy as they were in judging the symmetric Duchenne smiles as happy. Furthermore, they did not spend more time looking at the eyes or mouth regardless of types of smiles. While participants made more saccades between each side of the face for the asymmetric smiles than the symmetric ones, they judged the asymmetric smiles more often as really happy than not really happy. Thus, processing of these indices do not seem limited to perceptual-attentional difficulties as reflected in viewing behavior.

Sustained and transient attentional processes modulate neural predictors of memory encoding in consecutive time periods.

Memory formation is commonly thought to rely on brain activity following an event. Yet, recent research has shown that even brain activity previous to an event can predict later recollection (subsequent memory effect, SME). In order to investigate the attentional sources of the SME, event-related potentials (ERPs) elicited by task cues preceding target words were recorded in a switched task paradigm that was followed by a surprise recognition test. Stay trials, that is, those with the same task as the previous trial, were contrasted with switch trials, which included a task switch compared to the previous trial. The underlying assumption was that sustained attention would be dominant in stay trials and that transient attentional reconfiguration processes would be dominant in switch trials. To determine the SME, local and global statistics of scalp electric fields were used to identify differences between subsequently remembered and forgotten items. Results showed that the SME in stay trials occurred in a time window from 2 to 1 sec before target onset, whereas the SME in switch trials occurred subsequently, in a time window from 1 to 0 sec before target onset. Both SMEs showed a frontal negativity resembling the topography of previously reported effects, which suggests that sustained and transient attentional processes contribute to the prestimulus SME in consecutive time periods.