A5018244514- Cardiovascular Function and Risk Factors
- Advanced MRI Techniques and Applications
- Medical Image Segmentation Techniques
- Advanced Image Processing Techniques
- Cardiac Imaging and Diagnostics
- Image Processing Techniques and Applications
- Advanced Vision and Imaging
- Medical Imaging Techniques and Applications
- Cardiac Valve Diseases and Treatments
- Recommender Systems and Techniques
- Pulmonary Hypertension Research and Treatments
- Image and Signal Denoising Methods
- Advanced Neural Network Applications
- Advanced X-ray and CT Imaging
- Multimodal Machine Learning Applications
- Advanced Bandit Algorithms Research
- Mobile Crowdsensing and Crowdsourcing
- Coronary Interventions and Diagnostics
- Radiomics and Machine Learning in Medical Imaging
- Elasticity and Material Modeling
- Optical Coherence Tomography Applications
- Sentiment Analysis and Opinion Mining
- Video Analysis and Summarization
- Generative Adversarial Networks and Image Synthesis
- Cardiomyopathy and Myosin Studies
Yunnan Agricultural University
2025
ZTE (China)
2024-2025
State Key Laboratory of Mobile Networks and Mobile Multimedia Technology
2025
Twitter (United States)
2016-2020
Beijing Information Science & Technology University
2020
China University of Petroleum, Beijing
2018-2019
Imperial College London
2010-2018
MRC London Institute of Medical Sciences
2018
Northwestern Polytechnical University
2015-2017
National University of Singapore
2017
Despite the breakthroughs in accuracy and speed of single image super-resolution using faster deeper convolutional neural networks, one central problem remains largely unsolved: how do we recover finer texture details when super-resolve at large upscaling factors? The behavior optimization-based methods is principally driven by choice objective function. Recent work has focused on minimizing mean squared reconstruction error. resulting estimates have high peak signal-to-noise ratios, but...
Recently, several models based on deep neural networks have achieved great success in terms of both reconstruction accuracy and computational performance for single image super-resolution. In these methods, the low resolution (LR) input is upscaled to high (HR) space using a filter, commonly bicubic interpolation, before reconstruction. This means that super-resolution (SR) operation performed HR space. We demonstrate this sub-optimal adds complexity. paper, we present first convolutional...
Convolutional neural networks have enabled accurate image super-resolution in real-time. However, recent attempts to benefit from temporal correlations video been limited naive or inefficient architectures. In this paper, we introduce spatio-temporal sub-pixel convolution that effectively exploit redundancies and improve reconstruction accuracy while maintaining real-time speed. Specifically, discuss the use of early fusion, slow fusion 3D convolutions for joint processing multiple...
We propose a new approach to the problem of optimizing autoencoders for lossy image compression. New media formats, changing hardware technology, as well diverse requirements and content types create need compression algorithms which are more flexible than existing codecs. Autoencoders have potential address this need, but difficult optimize directly due inherent non-differentiabilty loss. here show that minimal changes loss sufficient train deep competitive with JPEG 2000 outperforming...
We consider the problem of face swapping in images, where an input identity is transformed into a target while preserving pose, facial expression and lighting. To perform this mapping, we use convolutional neural networks trained to capture appearance from unstructured collection his/her photographs. This approach enabled by framing terms style transfer, goal render image another one. Building on recent advances area, devise new loss function that enables network produce highly...
Despite the breakthroughs in accuracy and speed of single image super-resolution using faster deeper convolutional neural networks, one central problem remains largely unsolved: how do we recover finer texture details when super-resolve at large upscaling factors? The behavior optimization-based methods is principally driven by choice objective function. Recent work has focused on minimizing mean squared reconstruction error. resulting estimates have high peak signal-to-noise ratios, but...
Image super-resolution (SR) is an underdetermined inverse problem, where a large number of plausible high-resolution images can explain the same downsampled image. Most current single image SR methods use empirical risk minimisation, often with pixel-wise mean squared error (MSE) loss. However, outputs from such tend to be blurry, over-smoothed and generally appear implausible. A more desirable approach would employ Maximum Posteriori (MAP) inference, preferring solutions that always have...
We propose a new approach to the problem of optimizing autoencoders for lossy image compression. New media formats, changing hardware technology, as well diverse requirements and content types create need compression algorithms which are more flexible than existing codecs. Autoencoders have potential address this need, but difficult optimize directly due inherent non-differentiabilty loss. here show that minimal changes loss sufficient train deep competitive with JPEG 2000 outperforming...
Recently, several models based on deep neural networks have achieved great success in terms of both reconstruction accuracy and computational performance for single image super-resolution. In these methods, the low resolution (LR) input is upscaled to high (HR) space using a filter, commonly bicubic interpolation, before reconstruction. This means that super-resolution (SR) operation performed HR space. We demonstrate this sub-optimal adds complexity. paper, we present first convolutional...
Purpose To determine if patient survival and mechanisms of right ventricular failure in pulmonary hypertension could be predicted by using supervised machine learning three-dimensional patterns systolic cardiac motion. Materials Methods The study was approved a research ethics committee, participants gave written informed consent. Two hundred fifty-six patients (143 women; mean age ± standard deviation, 63 years 17) with newly diagnosed underwent magnetic resonance (MR) imaging, right-sided...
Late Gadolinium enhancement (LGE) cardiovascular magnetic resonance (CMR) imaging can be used to visualise regions of fibrosis and scarring in the left atrium (LA) myocardium. This important for treatment stratification patients with atrial fibrillation (AF) assessment after radio frequency catheter ablation (RFCA). In this paper we present a standardised evaluation benchmarking framework algorithms segmenting scar from LGE CMR images. The reported are response an open challenge that was put...
The most prominent problem associated with the deconvolution layer is presence of checkerboard artifacts in output images and dense labels. To combat this problem, smoothness constraints, post processing different architecture designs have been proposed. Odena et al. highlight three sources artifacts: overlap, random initialization loss functions. In note, we proposed an method for sub-pixel convolution known as NN resize. Compared to initialized schemes designed standard kernels, it free...
In this note, we want to focus on aspects related two questions most people asked us at CVPR about the network presented. Firstly, What is relationship between our proposed layer and deconvolution layer? And secondly, why are convolutions in low-resolution (LR) space a better choice? These key tried answer paper, but were not able go into as much depth clarity would have liked allowance. To these first discuss relationships forms of transposed convolution layer, sub-pixel convolutional...
An unresolved issue in patients with diastolic dysfunction is that the estimation of myocardial stiffness cannot be decoupled from residual active tension (AT) because impaired ventricular relaxation during diastole. To address this problem, paper presents a method for estimating mechanical parameters left ventricle (LV) cine and tagged MRI measurements LV cavity pressure recordings, separating passive constitutive properties AT. Dynamic C1-continuous meshes are automatically built anatomy...
Purpose: Cardiac computed tomography (CT) is widely used in clinical diagnosis of cardiovascular diseases. Whole heart segmentation (WHS) plays a vital role developing new applications cardiac CT. However, the shape and appearance can vary greatly across different scans, making automatic particularly challenging. The objective this work to develop evaluate multiatlas (MAS) scheme using atlas ranking selection algorithm for WHS CT data. Research on MAS strategies their influence performance...
In this paper, we present a novel technique based on nonrigid image registration for myocardial motion estimation using both untagged and 3-D tagged MR images. The aspect of our is its simultaneous usage complementary information from To estimate the within myocardium, register sequence images during cardiac cycle to set reference at end-diastole. similarity measure spatially weighted maximize utility addition, proposed approach integrates valve plane tracker adaptive incompressibility into...
Modulation recognition is a major task in many wireless communication systems including cognitive radio and signal reconnaissance. The diversification of modulation schemes the increased complexity channel environment put higher requirements on correct identification modulated signals. Deep learning (DL) considered as potential solution to solve these problems due superior big data processing classification capabilities. This paper proposes an efficient digital method based deep neural...
Cardiac phenotypes, such as left ventricular (LV) mass, demonstrate high heritability although most genes associated with these complex traits remain unidentified. Genome-wide association studies (GWAS) have relied on conventional 2D cardiovascular magnetic resonance (CMR) the gold-standard for phenotyping. However this technique is insensitive to regional variations in wall thickness which are often hypertrophy and require large cohorts reach significance. Here we test whether automated...