A5100748707- AI in cancer detection
- Radiomics and Machine Learning in Medical Imaging
- Digital Imaging for Blood Diseases
- Cell Image Analysis Techniques
- Advanced Image and Video Retrieval Techniques
- Speech and Audio Processing
- Human Pose and Action Recognition
- Advanced Image Processing Techniques
- Video Surveillance and Tracking Methods
- Wireless Communication Networks Research
- Image Retrieval and Classification Techniques
- Advanced Vision and Imaging
- Advanced Neural Network Applications
- Direction-of-Arrival Estimation Techniques
- Human Motion and Animation
- Single-cell and spatial transcriptomics
- UAV Applications and Optimization
- Video Coding and Compression Technologies
- Text and Document Classification Technologies
- Cervical Cancer and HPV Research
- Music and Audio Processing
- Forensic Anthropology and Bioarchaeology Studies
- Medical Image Segmentation Techniques
- Machine Learning and Data Classification
- Scientific and Engineering Research Topics
Tencent (China)
2019-2024
General Hospital of Shenyang Military Region
2024
Guizhou University
2023
Sichuan University of Science and Engineering
2023
Yibin University
2023
Xiangtan University
2021
Beijing Jiaotong University
2004-2010
Capital Normal University
2008
The large-scale whole-slide images (WSIs) facilitate the learning-based computational pathology methods. However, gigapixel size of WSIs makes it hard to train a conventional model directly. Current approaches typically adopt multiple-instance learning (MIL) tackle this problem. Among them, MIL combined with graph convolutional network (GCN) is significant branch, where sampled patches are regarded as nodes further discover their correlations. difficult build correspondence across from...
Background and aims Evaluation of the programmed cell death ligand‐1 (PD‐L1) combined positive score (CPS) is vital to predict efficacy immunotherapy in triple‐negative breast cancer (TNBC), but pathologists show substantial variability consistency accuracy interpretation. It great importance establish an objective effective method which highly repeatable. Methods We proposed a model deep learning‐based framework, at patch level incorporated analysis tissue region analysis, followed by...
Microbial communities, demonstrating dynamic changes in cadavers and the surroundings, provide invaluable insights for forensic investigations. Conventional methodologies microbiome sequencing data analysis face obstacles due to subjectivity inefficiency. Artificial Intelligence (AI) presents an efficient accurate tool, with ability autonomously process analyze high-throughput data, assimilate multi-omics encompassing metagenomics, transcriptomics, proteomics. This facilitates estimation of...
Text-guided motion synthesis aims to generate 3D human that not only precisely reflects the textual description but reveals details as much possible. Pioneering methods explore diffusion model for text-to-motion and obtain significant superiority. However, these conduct processes either on raw data distribution or low-dimensional latent space, which typically suffer from problem of modality inconsistency detail-scarce. To tackle this problem, we propose a novel Basic-to-Advanced Hierarchical...
We are concerned with the Calder\'on inverse inclusion problem, where one intends to recover shape of an inhomogeneous conductive embedded in a homogeneous conductivity by associated boundary measurements. consider highly challenging case single partial measurement, which constitutes long-standing open problem literature. It is shown several existing works that corner singularities can help resolve uniqueness and stability issues for this problem. In paper, we show singularity be relaxed...
The high-content image-based assay is commonly leveraged for identifying the phenotypic impact of genetic perturbations in biology field. However, a persistent issue remains unsolved during experiments: interferential technical noises caused by systematic errors (e.g., temperature, reagent concentration, and well location) are always mixed up with real biological signals, leading to misinterpretation any conclusion drawn. Here, we reported mean teacher-based deep learning model (DeepNoise)...
In the domain of 3D scene representation, Gaussian Splatting (3DGS) has emerged as a pivotal technology. However, its application to large-scale, high-resolution scenes (exceeding 4k$\times$4k pixels) is hindered by excessive computational requirements for managing large number Gaussians. Addressing this, we introduce 'EfficientGS', an advanced approach that optimizes 3DGS high-resolution, large-scale scenes. We analyze densification process in and identify areas over-proliferation. propose...
Generally, the conventional 2D-RAKE receivers are based on second-order statistics while assuming perfect array conditions. However, sensor response, location uncertainty, and use of sample can severely degrade performance processing. In this paper a cumulant-based receiver for synchronous CDMA system with decorrelator is presented. Employing signal processing technique proposed blind demonstrates excellent performance.
The state-of-the-art neural video codecs have outperformed the most sophisticated traditional in terms of RD performance certain cases. However, utilizing them for practical applications is still challenging two major reasons. 1) Cross-platform computational errors resulting from floating point operations can lead to inaccurate decoding bitstream. 2) high complexity encoding and process poses a challenge achieving real-time performance. In this paper, we propose cross-platform codec, which...
Under certain circumstances, advanced neural video codecs can surpass the most complex traditional in their rate-distortion (RD) performance. One of main reasons for high performance existing is use entropy model, which provide more accurate probability distribution estimations compressing latents. This also implies rigorous requirement that models running on different platforms should consistent estimations. However, cross-platform scenarios, usually yield inconsistent due to floating point...
A novel location algorithm based on geolocation and LANDMARC is proposed to improve the accuracy of Radio Frequency Identification (RFID) system in non-line-of-sight (NLOS) environment. We compare this with existing via two metrics: mean error cumulative distribution line-of-sight (LOS) NLOS environments, respectively. Numerical simulation results show that very simple, it gives higher environment little performance degradation compared LOS
Abstract Spatially resolved transcriptomics (SRT) has greatly expanded our understanding of the spatial patterns gene expression in histological tissue sections. However, most currently available platforms could not provide situ single-cell transcriptomics, limiting their biological applications. Here, to silico reconstruct SRT at resolution, we propose St2cell which combines deep learning-based frameworks with a novel convex quadratic programming (CQP)-based model. can thoroughly leverage...
The classification of nuclei in H&E-stained histopathological images is a fundamental step the quantitative analysis digital pathology. Most existing methods employ multi-class on detected nucleus instances, while annotation scale greatly limits their performance. Moreover, they often downplay contextual information surrounding instances that critical for classification. To explicitly provide to model, we design new structured input consisting content-rich image patch and target instance...
Labels are costly and sometimes unreliable. Noisy label learning, semi-supervised contrastive learning three different strategies for designing processes requiring less annotation cost. Semi-supervised have been recently demonstrated to improve that address datasets with noisy labels. Still, the inner connections between these fields as well potential combine their strengths together only started emerge. In this paper, we explore further ways advantages fuse them. Specifically, propose CSSL,...
Multi-object tracking (MOT) is a challenging vision task that aims to detect individual objects within single frame and associate them across multiple frames. Recent MOT approaches can be categorized into two-stage tracking-by-detection (TBD) methods one-stage joint detection (JDT) methods. Despite the success of these approaches, they also suffer from common problems, such as harmful global or local inconsistency, poor trade-off between robustness model complexity, lack flexibility in...
Mobile signal strength affects the deployment of IoT devices, and its distribution is often measured using UAVs designed reasonable routes equipped with measuring equipment. Traditional ant colony algorithms used in track planning can easily fall into local optimal solutions not suitable for large-scale tasks, an improved algorithm proposed solving problem. The characteristic that process updating pheromones, information release function changes number iterations introduced to avoid falling...