A5101491780- Voice and Speech Disorders
- Speech Recognition and Synthesis
- Medical Image Segmentation Techniques
- Advanced MRI Techniques and Applications
- Radiomics and Machine Learning in Medical Imaging
- Domain Adaptation and Few-Shot Learning
- Phonetics and Phonology Research
- Speech and Audio Processing
- Medical Imaging Techniques and Applications
- Cleft Lip and Palate Research
- Advanced X-ray and CT Imaging
- Cardiac Imaging and Diagnostics
- Advanced Neuroimaging Techniques and Applications
- Cancer-related molecular mechanisms research
- COVID-19 diagnosis using AI
- Generative Adversarial Networks and Image Synthesis
- Music and Audio Processing
- Brain Tumor Detection and Classification
- Advanced Neural Network Applications
- AI in cancer detection
- Salivary Gland Tumors Diagnosis and Treatment
- Fetal and Pediatric Neurological Disorders
- Advanced Image Processing Techniques
- Image and Signal Denoising Methods
- Traditional Chinese Medicine Studies
Gordon Center for Medical Imaging
2016-2024
Harvard University
2015-2024
Massachusetts General Hospital
2015-2024
East Carolina University
2023
Google (United States)
2023
University of Illinois Urbana-Champaign
2023
Novant Health
2023
Dana-Farber/Harvard Cancer Center
2023
University of Maryland, Baltimore
2011-2019
Johns Hopkins University
2012-2019
Multimodal image registration is a class of algorithms to find correspondence from different modalities. Since modalities do not exhibit the same characteristics, finding accurate still remains challenge. To deal with this, mutual information (MI)-based has been preferred choice as MI based on statistical relationship between both volumes be registered. However, some limitations. First, MI-based often fails when there are local intensity variations in volumes. Second, only considers...
Our aim in this study was to optimize and validate an adaptive denoising algorithm based on Block-Matching 3D, for reducing image noise improving assessment of left ventricular function from low-radiation dose coronary CTA. In paper, we describe the its validation, with CTA datasets 7 consecutive patients. We validated using a novel method, myocardial mass low-noise cardiac phase as reference standard, objective measurement noise. After denoising, were not statistically different by...
In this work, we propose an adversarial unsupervised domain adaptation (UDA) method under inherent conditional and label shifts, in which aim to align the distributions w.r.t. both p(x|y) p(y). Since labels are inaccessible a target domain, conventional UDA methods assume that p(y) is invariant across domains rely on aligning p(x) as alternative alignment. To address this, provide thorough theoretical empirical analysis of novel practical optimization scheme for UDA. Specifically, infer...
Computer-aided segmentation of cardiac images obtained by various modalities plays an important role and is a prerequisite for wide range applications facilitating the delineation anatomical regions interest. Numerous computerized methods have been developed to tackle this problem. Recent studies employ sophisticated techniques using available cues from anatomy such as geometry, visual appearance, prior knowledge. In addition, new minimization computational adopted with improved speed...
Sequential testing by coronary CT angiography (CTA) and myocardial perfusion SPECT (MPS) obtained on stand-alone scanners may be needed to diagnose artery disease in equivocal cases. We have developed an automated technique for MPS–CTA registration demonstrate its utility improved MPS quantification guiding the coregistered physiologic with anatomic CTA information. <b>Methods:</b> Automated of left ventricular (LV) surfaces trees was accomplished iterative minimization voxel differences...
Magnetic resonance images of the tongue have been used in both clinical studies and scientific research to reveal structure. In order extract different features its relation vocal tract, it is beneficial acquire three orthogonal image volumes-e.g., axial, sagittal, coronal volumes. maintain low noise high visual detail minimize blurred effect due involuntary motion artifacts, each set acquired with an in-plane resolution that much better than through-plane resolution. As a result, any one...
The human tongue has a complex architecture, consistent with its roles in eating, speaking and breathing. Tongue muscle architecture been depicted drawings photographs, but not quantified volumetrically. This paper aims to fill that gap by measuring the of for 14 people captured high-resolution 3D MRI volumes. results show structure, relationships variability among muscles, as well effects age, gender weight on volume. Since consists partially interdigitated we consider volumes two ways....
Tagged magnetic resonance imaging has been used for decades to observe and quantify motion strain of deforming tissue. It is challenging obtain 3-D estimates due a tradeoff between image slice density acquisition time. Typically, interpolation methods are either combine 2-D extracted from sparse acquisitions into or construct dense volume before registration applied. This paper proposes new phase-based estimation technique that first computes harmonic phase volumes interpolated tagged slices...
Semantic segmentation is a class of methods to classify each pixel in an image into semantic classes, which critical for autonomous vehicles and surgery systems. Cross-entropy (CE) loss-based deep neural networks (DNN) achieved great success w.r.t. the accuracy-based metrics, e.g., mean Intersection-over Union. However, CE loss has limitation that it ignores varying degrees severity pair-wise misclassified results. For instance, classifying car road much more terrible than recognizing as...
Multimodal MRI provides complementary and clinically relevant information to probe tissue condition characterize various diseases. However, it is often difficult acquire sufficiently many modalities from the same subject due limitations in study plans, while quantitative analysis still demanded. In this work, we propose a unified conditional disentanglement framework synthesize any arbitrary modality an input modality. Our hinges on cycleconstrained adversarial training approach, where can...
Recent advances in unsupervised domain adaptation (UDA) show that transferable prototypical learning presents a powerful means for class conditional alignment, which encourages the closeness of cross-domain centroids. However, inner-class compactness and underlying fine-grained subtype structure remained largely underexplored. In this work, we propose to adaptively carry out subtype-aware alignment by explicitly enforcing class-wise separation subtype-wise with intermediate pseudo labels....
Deep learning has great potential for accurate detection and classification of diseases with medical imaging data, but the performance is often limited by number training datasets memory requirements. In addition, many deep models are considered a "black-box," thereby limiting their adoption in clinical applications. To address this, we present successive subspace model, termed VoxelHop, Amyotrophic Lateral Sclerosis (ALS) using T2-weighted structural MRI data. Compared popular convolutional...
In this work, we propose a domain generalization (DG) approach to learn on several labeled source domains and transfer knowledge target that is inaccessible in training. Considering the inherent conditional label shifts, would expect alignment of p(x|y) p(y). However, widely used invariant feature learning (IFL) methods relies aligning marginal concept shift w.r.t. p(x), which rests an unrealistic assumption p(y) across domains. We thereby novel variational Bayesian inference framework...
Unsupervised domain adaptation (UDA) between two significantly disparate domains to learn high-level semantic alignment is a crucial yet challenging task. To this end, in work, we propose exploiting low-level edge information facilitate the as precursor task, which has small cross-domain gap, compared with segmentation. The precise contour then provides spatial guide adaptation. More specifically, multi-task framework contouring network along segmentation network, takes both magnetic...