Short videos follow the trend of creation, leading to a proliferation of homogenized video content. Textual overlays such as titles in short videos often reflect semantic homogeneity. This phenomenon manifests not only in the syntactic structure and overall thematic expression, but also in local semantic elements such as words and phrases. Based on information retrieval techniques, we propose a text semantic homogenization recognition network (TSHR-Net) for short video title overlays. The framework comprises key components: (1) a dynamic semantic representation that incorporates contextual information of title overlays using RoBERTa pre-trained word embeddings; (2) a dual-path semantic parser that integrates global semantics via BiLSTM-Attention and local semantics via multi-scale TextCNN; (3) a ranking loss optimization is designed to measure cosine similarity between semantic features, thereby improving homogenization recognition accuracy. Experimental results show that our TSHR-Net achieves the competitive performance in Chinese text semantic homogenization recognition, with ρ and ρ X,Y reaching 80.16% and 78.23% on LCQMC, 81.05% and 80.19% on STS-B(ZH), and 92.47% and 92.19% on our self-built BJUT-HCD. The model also exhibits generalization ability in English, attaining 79.68% and 79.89% on STS-B(EN), and 73.56% and 72.21% on SICK dataset, respectively.

Automated segmentation of mediastinal neoplasms with preoperative computed tomography (CT) scans is critical for clinical diagnosis. Although convolutional neural networks (CNNs) have proven effective in medical imaging segmentation, the segmentation of mediastinal neoplasms, which vary greatly in shape, size, and texture, presents a unique challenge due to the inherent local focus of convolution operations. To address this limitation, we propose a confidence-enhanced semi-supervised learning framework for mediastinal neoplasm segmentation. Specifically, we introduce a confidence-enhanced module that improves segmentation accuracy over indistinct tumor boundaries by assessing and excluding unreliable predictions simultaneously, which can greatly enhance the efficiency of exploiting unlabeled data. In addition, we implement an iterative learning strategy designed to continuously refine prediction reliability estimates throughout the training process, ensuring more precise confidence assessments. Quantitative analysis on a real-world dataset demonstrates that our model significantly improves the performance by leveraging unlabeled data, surpassing existing semi-supervised segmentation benchmarks. Finally, to promote more efficient academic communication, the analysis code is available at https://github.com/fxiaotong432/CEDS

Automatic segmentation and classification of thymic tumors based on preoperative CT scans are critical for clinical diagnosis. However, significant variability in the shape, size, and texture of thymic tumors, along with their blurred boundaries and complex pathological features, poses substantial challenges to automated recognition. This study focuses on two key objectives: (1) semantic segmentation at the pixel level of thymic tumors on CT images and (2) identification of high-risk thymic carcinoma. To address the above challenges, we propose a multiview attention-guided network (MAG-Net), a novel multitask learning framework guided by attention mechanisms. The model simultaneously takes 3D subvolumes of equal size extracted from axial, coronal, and sagittal views of the CT scan as input and fuses features across multiple views under the guidance of attention mechanisms. Moreover, we introduce a segmentation-classification prior attention (SCPA) module that embeds spatial location cues from segmentation into feature learning for classification. Extensive experiments conducted on both our own collected data set and public data sets demonstrate the effectiveness of the proposed method, achieving a dice coefficient of 90.54% for the segmentation task and an AUC of nearly 0.9 for classification. To facilitate further research, the analysis code is available on https://github.com/weixuxuxu/MAGNet.

Short videos are one of the most popular social media in the world, triggering a proliferation of copycat creations leading to homogenized video content, with visual and textual homogenization being the most prevalent. Unlike near-duplicate video retrieval, which relies on visual appearance similarity, homogenization recognition emphasizes identifying videos with similar semantic units. Short videos exhibit multimodal features, in which there is a many-to-many mapping relationship between visual and text elements, and the two modalities are relatively independent and semantically correlated. Therefore, cross-modal semantic correlation needs to be explored and established to achieve homogenization recognition of short videos. Based on the idea of divide-and-conquer and joint processing, we propose a cross-modal tri-semantic correlation-CLIP (CS 3 C-CLIP) for short video homogenization recognition. First, visual and text features in the shared subspace are extracted using the contrastive language-image pre-training visual-text dual encoder. Then, features at the patch, frame, and video levels are generated using the patch selection module and the temporal encoder, while the word-level and sentence-level features are respectively derived from text features and [EOS] token. After establishing cross-modal tri-semantic correlations by constructing a triple semantic (i.e., video-sentence, frame-sentence, and patch-word) correlation, homogenized short videos are recognized by measuring the aggregated cross-modal similarity between pairs of short videos. Experimental results on three publicly available datasets demonstrate that our CS 3 C-CLIP outperforms state-of-the-art methods, achieving 85.7% R@1 and 94.4% R@5 on self-built BJUT-HCD, 49.4% R@1 and 74.6% R@5 on MSR-VTT, and 49.8% R@1 and 78.1% R@5 on MSVD, respectively.

This paper investigates dynamic spectrum access (DSA) for cognitive satellite networks (CSNs), where a non-geostationary orbit (NGSO) satellite acting as a secondary user (SU) shares a segment of spectrum licensed to the primary user (PU) geostationary (GSO) satellite system. Considering the influence of spectrum sensing errors on spectrum sharing, a new problem about joint channel selection and power control is formulated as a sequential decision-making process, to maximize a long-term throughput of the NGSO under an interference constraint for the GSO system. Due to the imperfect spectrum information, we employ deep reinforcement learning (DRL) with a double deep Q-learning neural network to solve the problem by learning the set of DSA policies with stabilized convergence. We consider three cases to evaluate the performance of the proposed algorithm and the simulation results show that the throughput and the average transmission power of NGSO can converge quickly in a dynamic spectrum environment. Besides, the spectrum utilization can be maximized with meeting the transmission power constraint proposed by GSO for NGSO.

MOTIVATION: Microsatellite instability (MSI) is a promising biomarker for cancer prognosis and chemosensitivity. Techniques are rapidly evolving for the detection of MSI from tumor-normal paired or tumor-only sequencing data. However, tumor tissues are often insufficient, unavailable, or otherwise difficult to procure. Increasing clinical evidence indicates the enormous potential of plasma circulating cell-free DNA (cfNDA) technology as a noninvasive MSI detection approach. RESULTS: We developed MSIsensor-ct, a bioinformatics tool based on a machine learning protocol, dedicated to detecting MSI status using cfDNA sequencing data with a potential stable MSIscore threshold of 20%. Evaluation of MSIsensor-ct on independent testing datasets with various levels of circulating tumor DNA (ctDNA) and sequencing depth showed 100% accuracy within the limit of detection (LOD) of 0.05% ctDNA content. MSIsensor-ct requires only BAM files as input, rendering it user-friendly and readily integrated into next generation sequencing (NGS) analysis pipelines. AVAILABILITY: MSIsensor-ct is freely available at https://github.com/niu-lab/MSIsensor-ct. SUPPLEMENTARY INFORMATION: Supplementary data are available at Briefings in Bioinformatics online.

Next-generation sequencing (NGS) technology has revolutionised human cancer research, particularly via detection of genomic variants with its ultra-high-throughput sequencing and increasing affordability. However, the inundation of rich cancer genomics data has resulted in significant challenges in its exploration and translation into biological insights. One of the difficulties in cancer genome sequencing is software selection. Currently, multiple tools are widely used to process NGS data in four stages: raw sequence data pre-processing and quality control (QC), sequence alignment, variant calling and annotation and visualisation. However, the differences between these NGS tools, including their installation, merits, drawbacks and application, have not been fully appreciated. Therefore, a systematic review of the functionality and performance of NGS tools is required to provide cancer researchers with guidance on software and strategy selection. Another challenge is the multidimensional QC of sequencing data because QC can not only report varied sequence data characteristics but also reveal deviations in diverse features and is essential for a meaningful and successful study. However, monitoring of QC metrics in specific steps including alignment and variant calling is neglected in certain pipelines such as the 'Best Practices Workflows' in GATK. In this review, we investigated the most widely used software for the fundamental analysis and QC of cancer genome sequencing data and provided instructions for selecting the most appropriate software and pipelines to ensure precise and efficient conclusions. We further discussed the prospects and new research directions for cancer genomics.

Internal tandem duplication (ITD) of FMS-like tyrosine kinase 3 (FLT3-ITD) constitutes an independent indicator of poor prognosis in acute myeloid leukaemia (AML). AML with FLT3-ITD usually presents with poor treatment outcomes, high recurrence rate and short overall survival. Currently, polymerase chain reaction and capillary electrophoresis are widely adopted for the clinical detection of FLT3-ITD, whereas the length and mutation frequency of ITD are evaluated using fragment analysis. With the development of sequencing technology and the high incidence of FLT3-ITD mutations, a multitude of bioinformatics tools and pipelines have been developed to detect FLT3-ITD using next-generation sequencing data. However, systematic comparison and evaluation of the methods or software have not been performed. In this study, we provided a comprehensive review of the principles, functionality and limitations of the existing methods for detecting FLT3-ITD. We further compared the qualitative and quantitative detection capabilities of six representative tools using simulated and biological data. Our results will provide practical guidance for researchers and clinicians to select the appropriate FLT3-ITD detection tools and highlight the direction of future developments in this field. Availability: A Docker image with several programs pre-installed is available at https://github.com/niu-lab/docker-flt3-itd to facilitate the application of FLT3-ITD detection tools.

Conventional neural generative models tend to generate safe and generic responses which have little connection with previous utterances semantically and would disengage users in a dialog system.To generate relevant responses, we propose a method that employs two types of constraints -topical constraint and semantic constraint.Under the hypothesis that a response and its context have higher relevance when they share the same topics, the topical constraint encourages the topics of a response to match its context by conditioning response decoding on topic words' embeddings.The semantic constraint, which encourages a response to be semantically related to its context by regularizing the decoding objective function with semantic distance, is proposed.Optimal transport is applied to compute a weighted semantic distance between the representation of a response and the context.Generated responses are evaluated by automatic metrics, as well as human judgment, showing that the proposed method can generate more topic-relevant and content-rich responses than conventional models.

The dynamic load on the bridge is normally generated by the traffic flow. It is therefore the acquisition of spatiotemporal information for vehicles driving on a bridge is of great significance to assess bridge structures. In this paper, we proposed an improved deep learning network, which is inspired from YOLOv4 (You only look once) network structure. The transfer learning method is applied in training, and designated nine sets of anchor values are obtained by the K-means clustering. The Soft-NMS (Non-Maximum Suppression) algorithm is fused to improve the detection effect of overlapping targets. At the same time, the SENet (Squeeze-and- Excitation Networks) is used to assign weights to the features of each channel to learn the correlation between different channels. Data set are established with video clips cut from the surveillance system currently used on a long span cable-stayed bridge in China. The proposed method is compared with SSD (Single Shot MultiBox Detector), YOLOv3 and YOLOv4 algorithms. Experimental proves that the proposed method can detect vehicle information more accurately, efficiently and stably. The result could extend to other bridge monitoring applications in future.

This paper proposes a distributed power control scheme based on the constrained model predictive control (MPC) for the underlay cognitive satellite terrestrial networks (CSTNs), where the primary satellite communication network coexists with the secondary terrestrial mobile network. We model this power control problem as a closed-loop dynamic control system with the inner loop and outer loop. On the basis of combining target power control (TPC) algorithm in the inner loop and tracking of flexible target signal to interference plus noise ratio (SINR) in the outer loop, we develop a corresponding state space expression of the problem where the fluctuation of each channel power gain is formulated as the exogenous disturbance input so that we do not need the accurate instantaneous channel state information (CSI). Then we design a SINR regulator in the outer loop, which is a constrained model predicted controller with rolling optimal operation subject to the interference temperature constraint obtained by calculating a linear matrix inequality. Finally, we obtain our constrained model predictive power control algorithm. In contrast to the previous static power control schemes based on the optimization theory that highly depend on the known instantaneous CSI and large signalling exchanges, the proposed scheme only needs locally measured information and outdate feedbacks. The performance of the proposed algorithm is shown to be effective through computer simulations.

Considering a random time-varying channel model, we propose a decentralized power allocation (PA) scheme based on ℋ∞control theory for a cognitive radio network (CRN) under dynamic formulation by state space model with exogenous input. In this state space model, we transform the interference temperature (IT) constraint and the target signal to interference plus noise ratio (SINR) tracking to a weighted control performance index. We design a ℋ∞controller to make the index minimum to obtain a reasonable target SINR. In the PA scheme, each active secondary user (SU) controls its transmit power related with its instantaneous SINR to track the target SINR. Simulation results show that the proposed strategy using ℋ∞controller is effective and valid for the SINR and IT requirements of both SUs and primary user (PU).

The author brings forward a general object network management platform based on MOF (meta object facility) in the paper. The platform is based on MOF model gave out by OMG, and adopts the modeling idea of model driver and MOF meta-meta model (namely the model of the meta model). In addition, the model driver idea is applied in the software development process; software development is based on a high layer model abstract. Those objects which need to be managed are abstracted layer by layer and make an independent model in different layer. So, software development process is switched into objects modeling process, and the model structure of whole network management platform is built up step by step. The object modeling is not affected by realization technology and platform. It could realize the upgrade and expansion of network management software quickly, solve the problems of flexibility, diversity and changeability that is currently met by network management software, rapidly adapt the demands of network continuously expansion and operation enlarge, improve the reuse ability of models.

Network data collection is the important part in network management, and the system design involves a lot of concurrency and communication problems. Therefore a formal approach was presented to the formal analysis of network data collection, based on Hoare's communication sequential process (CSP) and some theoretical results of network formal reaction. Then the formal analysis to network data collection was carried through expanding CSP. A solid mathematics foundation of the correctness test was provided for system design through the precise formal description for network data collection.

This paper describes a practical approach of network management software for telecom domain based on model-driven approach and CSCW technology. A modeling tool, customizing tool and telecom network object framework have been developed. The modeling tool provides an environment on which the modeler can model domain entities and relationships, and the assemblage rules of models and the domain framework components are made by customizing tool using relationship paths. The domain framework cooperates with domain models to implement concrete applications by customizing process, consequently the domain specific application software is generated