A5083630052- Radiomics and Machine Learning in Medical Imaging
- AI in cancer detection
- Privacy-Preserving Technologies in Data
- Image Enhancement Techniques
- Medical Imaging Techniques and Applications
- Human Pose and Action Recognition
- Advanced Vision and Imaging
- Photoacoustic and Ultrasonic Imaging
- Machine Learning in Healthcare
- Advanced MRI Techniques and Applications
- Multimodal Machine Learning Applications
- Advanced Fluorescence Microscopy Techniques
- Medical Image Segmentation Techniques
- Video Surveillance and Tracking Methods
- Medical Imaging and Analysis
- Domain Adaptation and Few-Shot Learning
- Advanced Image and Video Retrieval Techniques
- Optical Imaging and Spectroscopy Techniques
- MRI in cancer diagnosis
- Color Science and Applications
- Artificial Intelligence in Healthcare
- Brain Tumor Detection and Classification
- Stochastic Gradient Optimization Techniques
- Cryptography and Data Security
- Retinal Imaging and Analysis
University of Hong Kong
2023-2025
Chinese University of Hong Kong
2025
University of North Carolina at Chapel Hill
2018-2023
Stanford University
2020-2023
Imaging Center
2020-2021
Chinese Academy of Sciences
2015-2020
Shenyang Institute of Automation
2015-2020
University of Chinese Academy of Sciences
2015-2017
City University of Hong Kong
2016-2017
Shadow removal is a challenging task as it requires the detection/annotation of shadows well semantic understanding scene. In this paper, we propose an automatic and end-to-end deep neural network (DeshadowNet) to tackle these problems in unified manner. DeshadowNet designed with multi-context architecture, where output shadow matte predicted by embedding information from three different perspectives. The first global extracts features view. Two levels are derived transferred two parallel...
Numerous efforts have been made to design different low level saliency cues for the RGBD detection, such as color or depth contrast features, background and compactness priors. However, how these interact with each other incorporate effectively generate a master map remain challenging problem. In this paper, we new convolutional neural network (CNN) fuse into hierarchical features automatically detecting salient objects in images. existing works that directly feed raw image pixels CNN,...
Federated learning is an emerging research paradigm enabling collaborative training of machine models among different organizations while keeping data private at each institution. Despite recent progress, there remain fundamental challenges such as the lack convergence and potential for catastrophic forgetting across real-world heterogeneous devices. In this paper, we demonstrate that self-attention-based architectures (e.g., Transformers) are more robust to distribution shifts hence improve...
The collection and curation of large-scale medical datasets from multiple institutions is essential for training accurate deep learning models, but privacy concerns often hinder data sharing. Federated (FL) a promising solution that enables privacy-preserving collaborative among different institutions, it generally suffers performance deterioration due to heterogeneous distributions lack quality labeled data. In this paper, we present robust label-efficient self-supervised FL framework image...
3D hand pose tracking/estimation will be very important in the next generation of human-computer interaction. Most currently available algorithms rely on low-cost active depth sensors. However, these sensors can easily interfered by other sources and require relatively high power consumption. As a result, they are not suitable for outdoor environments mobile devices. This paper aims at tracking/estimating poses using passive stereo which avoids limitations. A benchmark with 18,000 image...
Accurate lesion segmentation based on endoscopy images is a fundamental task for the automated diagnosis of gastrointestinal tract (GI Tract) diseases. Previous studies usually use hand-crafted features representing images, while feature definition and are treated as two standalone tasks. Due to possible heterogeneity between models, these methods often result in sub-optimal performance. Several fully convolutional networks have been recently developed jointly perform learning model training...
Significance Three-dimensional microscopy in the NIR-II window (1,000 to 1,700 nm) allows noninvasive deep-tissue optical sectioning of live mammals with high spatiotemporal resolution due suppressed light scattering and reduced tissue autofluorescence. Herein, we present a structured-illumination light-sheet (NIR-II SIM) both excitation emission wavelengths NIR-IIb (1,500 nm). Integrating structured illumination into further diminished out-of-focus background improved spatial resolution,...
In vivo fluorescence/luminescence imaging in the near-infrared-IIb (NIR-IIb, 1,500 to 1,700 nm) window under <1,000 nm excitation can afford subcentimeter depth without any tissue autofluorescence, promising high-precision intraoperative navigation clinic. Here, we developed a compact imager for concurrent visible photographic and NIR-II (1,000 3,000 fluorescence preclinical image-guided surgery. Biocompatible erbium-based rare-earth nanoparticles (ErNPs) with bright down-conversion...
Federated learning is an emerging research paradigm for enabling collaboratively training deep models without sharing patient data. However, the data from different institutions are usually heterogeneous across institutions, which may reduce performance of trained using federated learning. In this study, we propose a novel heterogeneity-aware method, SplitAVG, to overcome drops heterogeneity in Unlike previous methods that require complex heuristic or hyper parameter tuning, our SplitAVG...
In this paper we establish a long-term 3D hand pose tracking benchmark <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup> . It contains 18,000 stereo image pairs as well the ground-truth positions of palm and finger joints from different scenarios. Meanwhile, to accurately segment images, propose novel stereo-based segmentation depth estimation algorithm specially tailored for here. The experiments indicate effectiveness proposed by...
Human-Object Interaction (HOI) detection aims to infer interactions between humans and objects, it is very important for scene analysis understanding. The existing methods usually focus on exploring instance-level (e.g., object appearance) or interaction-level action semantic) features conduct interaction prediction. However, most of these only consider the self-triplet feature aggregation, which may lead learning ambiguity without cross-triplet context exchange. In this paper, from both...
Shadow removal, which aims to restore the illumination in shadow regions, is challenging due diversity of shadows terms location, intensity, shape, and size. Different from most multi-task methods, design elaborate multi-branch or multi-stage structures for better we introduce feature decomposition learn representations. Specifically, propose a single-stage decoupled network (DMTN) explicitly decomposed features matte estimation, image reconstruction. First, several coarse-to-fine...
Multi-head attention (MA), which allows the model to jointly attend crucial information from diverse representation subspaces through its heads, has yielded remarkable achievement in image captioning. However, there is no explicit mechanism ensure MA attends appropriate positions subspaces, resulting overfocused for each head and redundancy between heads. In this paper, we propose a novel Intra- Inter-Head Orthogonal Attention (I2OA) efficiently improve captioning by introducing concise...
Segmenting breast tumors from dynamic contrast-enhanced magnetic resonance (DCE-MR) images is a critical step for early detection and diagnosis of cancer. However, variable shapes sizes tumors, as well inhomogeneous background, make it challenging to accurately segment in DCE-MR images. Therefore, this article, we propose novel tumor-sensitive synthesis module demonstrate its usage after being integrated with tumor segmentation. To suppress false-positive segmentation similar contrast...
Brain magnetic resonance imaging (MRI) provides detailed soft tissue contrasts that are critical for disease diagnosis and neuroscience research. Higher MRI resolution typically comes at the cost of signal-to-noise ratio (SNR) contrast, particularly more common 3 Tesla (3T) scanners. At ultra-high field strength, 7 (7T) allows higher with greater contrast SNR. However, prohibitively high costs 7T scanners deter their widespread adoption in clinical research centers. To obtain higher-quality...