A5101476695- Advanced Neural Network Applications
- Advanced X-ray and CT Imaging
- Radiomics and Machine Learning in Medical Imaging
- Medical Image Segmentation Techniques
- Medical Imaging and Analysis
- AI in cancer detection
- COVID-19 diagnosis using AI
- Medical Imaging Techniques and Applications
- Renal and Vascular Pathologies
- Renal cell carcinoma treatment
- Cardiac Imaging and Diagnostics
- MRI in cancer diagnosis
- Multimodal Machine Learning Applications
- Brain Tumor Detection and Classification
- Human Pose and Action Recognition
- Domain Adaptation and Few-Shot Learning
- Advanced MRI Techniques and Applications
- Retinal Imaging and Analysis
- Image Enhancement Techniques
- Neonatal Respiratory Health Research
- Smart Agriculture and AI
- Coronary Interventions and Diagnostics
- Colorectal Cancer Screening and Detection
- Nanoporous metals and alloys
- Frequency Control in Power Systems
Southeast University
2019-2025
Guilin University of Electronic Technology
2022-2025
Guilin University
2025
Western University
2023-2024
Zhongshan Hospital
2019-2023
Fudan University
2019-2023
Ningxia University
2023
Second Affiliated Hospital of Zhejiang University
2022
Huazhong Agricultural University
2022
South China Normal University
2022
Medical image segmentation is a critical component in clinical practice, facilitating accurate diagnosis, treatment planning, and disease monitoring. However, current methods predominantly rely on customized models, which exhibit limited generality across diverse tasks. In this study, we present MedSAM, the inaugural foundation model designed for universal medical segmentation. Harnessing power of meticulously curated dataset comprising over one million images, MedSAM not only outperforms...
Cardiovascular disease is a major cause of death worldwide. Computed Tomography Coronary Angiography (CTCA) non-invasive method used to evaluate coronary artery disease, as well evaluating and reconstructing heart vessel structures. Reconstructed models have wide array for educational, training research applications such the study diseased non-diseased anatomy, machine learning based risk prediction in-silico in-vitro testing medical devices. However, arteries are difficult image due their...
The nature of thick-slice scanning causes severe inter-slice discontinuities 3D medical images, and the vanilla 2D/3D convolutional neural networks (CNNs) fail to represent sparse information dense intra-slice in a balanced way, leading underfitting features (for 2D CNNs) overfitting noise from long-range slices CNNs). In this work, novel mesh network (MNet) is proposed balance spatial representation inter axes via learning. 1) Our MNet latently fuses plenty processes by embedding...
Learning inter-image similarity is crucial for 3D medical images self-supervised pre-training, due to their sharing of numerous same semantic regions. However, the lack prior in metrics and semantic-independent variation make it challenging get a reliable measurement similarity, hindering learning consistent representation semantics. We investigate problem this task, i.e., between clustering effect features. propose novel visual paradigm, Geometric Visual Similarity Learning, which embeds...
Deformable medical image registration estimates corresponding deformation to align the regions of interest (ROIs) two images a same spatial coordinate system. However, recent unsupervised models only have correspondence ability without perception, making misalignment on blurred anatomies and distortion task-unconcerned backgrounds. Label-constrained (LC) embed perception via labels, but lack texture constraints in labels expensive labeling costs causes internal ROIs overfitted perception. We...
The natural liver extracellular matrix (ECM) achieved by decellularization holds great potential in the fields of tissue engineering and regenerative medicine. Additionally, use crosslinking agents on ECM to stabilize its ultrastructure enhance scaffold durability is gaining interest engineering. objective this study was compare properties porcine crosslinked with different (glutaraldehyde, genipin, quercetin) find best strategy for producing a decellularized optimal stable characteristics...
In this work, we address the task of few-shot medical image segmentation (MIS) with a novel proposed framework based on learning registration to learn (LRLS) paradigm. To cope limitations lack authenticity, diversity, and robustness in existing LRLS frameworks, propose better (BRBS) three main contributions that are experimentally shown have substantial practical merit. First, improve authenticity registration-based generation program knowledge consistency constraint strategy constrains...
To study the respective peripheral and systemic mechanisms of action dexmedetomidine, as adjuvant to regional anesthesia, we compared dexmedetomidine added ropivacaine for mid-forearm nerve blocks, either systemic-only a control with no dexmedetomidine.Sixty patients undergoing hand surgery were randomly divided into three groups (n = 20 per group). Each group underwent triple-nerve (median, radial ulnar) blocks 0.75% ropivacaine. In DexP group, 60 µg anesthetic mixture, while in DexIV they...
Domain adaptation (DA) for cardiac ultrasound image segmentation is clinically significant and valuable. However, previous domain methods are prone to be affected by the incomplete pseudo-label low-quality target source images. Human-centric has great advantages of human cognitive guidance help model adapt reduce reliance on labels. Doctor gaze trajectories contains a large amount cross-domain guidance. To leverage information cognition guiding adaptation, we propose gaze-assisted...
Dense contrastive representation learning (DCRL) has greatly improved the efficiency for image dense prediction tasks, showing its great potential to reduce large costs of medical collection and annotation. However, properties images make unreliable correspondence discovery, bringing an open problem large-scale false positive negative (FP&N) pairs in DCRL. In this paper, we propose GEoMetric vIsual deNse sImilarity (GEMINI) which embeds homeomorphism prior DCRL enables a reliable discovery...
MXene, characterized by its unique layered structure, shows great promise as an electrode material for energy storage devices. MXene‐based materials, with their excellent metallic conductivity, high packing density, and large specific surface area, efficiently store high‐rate Faradaic pseudocapacitive energy. This study successfully synthesized a NiS2/Mxene composite using self‐assembly method. The structural design significantly increases the material’s area active sites while effectively...
Image captioning is a cross-task of computer vision and natural language processing, aiming to describe image content in language. Existing methods still have deficiencies modeling the spatial location semantic correlation between regions. Furthermore, these often exhibit insufficient interaction features text features. To address issues, we propose Linformer-based method, Dense Memory Linformer for Captioning (DMFormer), which has lower time space complexity than traditional Transformer...