A5085020647- Cardiac Valve Diseases and Treatments
- Cardiovascular Function and Risk Factors
- Cardiac Imaging and Diagnostics
- Advanced MRI Techniques and Applications
- Elasticity and Material Modeling
- Advanced Neural Network Applications
- Medical Image Segmentation Techniques
- Platelet Disorders and Treatments
- Erythrocyte Function and Pathophysiology
- Ultrasound Imaging and Elastography
- Medical Imaging and Analysis
- Medical Imaging Techniques and Applications
- Coronary Interventions and Diagnostics
- Single-cell and spatial transcriptomics
Yale University
2018-2024
Characterizing left ventricular deformation and strain using 3D+time echocardiography provides useful insights into cardiac function can be used to detect localize myocardial injury. To achieve this, it is imperative obtain accurate motion estimates of the ventricle. In many analysis pipelines, this step often accompanied by a separate segmentation step; however, recent works have shown both tasks highly related complementary when optimized jointly. work, we present multi-task learning...
Accurate interpretation and analysis of echocardiography is important in assessing cardiovascular health. However, motion tracking often relies on accurate segmentation the myocardium, which can be difficult to obtain due inherent ultrasound properties. In order address this limitation, we propose a semi-supervised joint learning network that exploits overlapping features segmentation. The simultaneously trains two branches: one for Each branch learns extract relevant their respective tasks...
Reliable motion estimation and strain analysis using 3D+ time echocardiography (4DE) for localization characterization of myocardial injury is valuable early detection targeted interventions. However, difficult due to the low-SNR that stems from inherent image properties 4DE, intelligent regularization critical producing reliable estimates. In this work, we incorporated notion domain adaptation into a supervised neural network framework. We first propose semi-supervised Multi-Layered...
Accurate motion tracking of the left ventricle is critical in detecting wall abnormalities heart after an injury such as a myocardial infarction. We propose unsupervised framework with physiological constraints to learn dense displacement fields between sequential pairs 2-D B-mode echocardiography images. Current deep-learning motion-tracking algorithms require large amounts data provide ground-truth, which difficult obtain for vivo datasets (such patient and animal studies), or are...
Abstract Single-cell assays have enriched our understanding of hematopoiesis and, more generally, stem and progenitor cell biology. However, these single-end-point approaches provide only a static snapshot the state cell. To observe measure dynamic changes that may instruct fate, we developed an approach for examining hematopoietic fate specification using long-term (> 7-day) single-cell time-lapse imaging up to 13 generations with in situ fluorescence staining primary human progenitors...
Speckle tracking based on block matching is the most common method for multi-dimensional motion estimation in ultrasound elasticity imaging. Extension of two-dimensional (2-D) methods to three dimensions (3-D) has been problematic because large computational load 3-D tracking, as well performance issues related low frame (volume) rates images. To address both these problems, we have developed an efficient two-pass suited cardiac PatchMatch, originally image editing, adapted provide...
Reliable motion estimation and strain analysis using 3D+time echocardiography (4DE) for localization characterization of myocardial injury is valuable early detection targeted interventions. However, difficult due to the low-SNR that stems from inherent image properties 4DE, intelligent regularization critical producing reliable estimates. In this work, we incorporated notion domain adaptation into a supervised neural network framework. We first propose an unsupervised autoencoder with...
Accurate motion estimation and segmentation of the left ventricle from medical images are important tasks for quantitative evaluation cardiovascular health. Echocardiography offers a cost-efficient non-invasive modality examining heart, but provides additional challenges automated analyses due to low signal-to-noise ratio inherent in ultrasound imaging. In this work, we propose shape regularized convolutional neural network estimating dense displacement fields between sequential 3D B-mode...
The accurate quantification of left ventricular (LV) deformation/strain shows significant promise for quantitatively assessing cardiac function use in diagnosis and therapy planning (Jasaityte et al., 2013). However, estimation the displacement myocardial tissue hence LV strain has been challenging due to a variety issues, including those related deriving tracking tokens from images following locations over entire cycle. In this work, we propose point matching scheme where correspondences...