Accelerating whole-heart 3D T2 mapping: Impact of undersampling strategies and reconstruction techniques

Purpose We aim to determine an advantageous approach for the acceleration of high spatial resolution 3D cardiac T2 relaxometry data by comparing the performance of different undersampling patterns and reconstruction methods over a range of acceleration rates. Methods Multi-volume 3D high-resolution cardiac images were acquired fully and undersampled retrospectively using 1) optimal CAIPIRINHA and 2) a variable density random (VDR) sampling. Data were reconstructed using 1) multi-volume sensitivity encoding (SENSE), 2) joint-sparsity SENSE and 3) model-based SENSE. Four metrics were calculated on 3 naïve swine and 8 normal human subjects over a whole left-ventricular region of interest: root-mean-square error (RMSE) of image signal intensity, RMSE of T2, the bias of mean T2, and standard deviation (SD) of T2. Fully sampled data and volume-by-volume SENSE with standard equally spaced undersampling were used as references. The Jaccard index calculated from one swine with acute myocardial infarction (MI) was used to demonstrate preservation of segmentation of edematous tissues with elevated T2. Results In naïve swine and normal human subjects, all methods had similar performance when the net reduction factor (Rnet) <2.5. VDR sampling with model-based SENSE showed the lowest RMSEs (10.5%-14.2%) and SDs (+1.7–2.4 ms) of T2 when Rnet>2.5, while VDR sampling with the joint-sparsity SENSE had the lowest bias of mean T2 (0.0–1.1ms) when Rnet>3. The RMSEs of parametric T2 values (9.2%-24.6%) were larger than for image signal intensities (5.2%-18.4%). In the swine with MI, VDR sampling with either joint-sparsity or model-based SENSE showed consistently higher Jaccard index for all Rnet (0.71–0.50) than volume-by-volume SENSE (0.68–0.30). Conclusions Retrospective exploration of undersampling and reconstruction in 3D whole-heart T2 parametric mapping revealed that maps were more sensitive to undersampling than images, presenting a more stringent limiting factor on Rnet. The combination of VDR sampling patterns with model-based or joint-sparsity SENSE reconstructions were more robust for Rnet>3.

Double Regression-Based Sparse Unmixing for Hyperspectral Images

Sparse unmixing has attracted widespread attention from researchers, and many effective unmixing algorithms have been proposed in recent years. However, most algorithms improve the unmixing accuracy at the cost of large calculations. Higher unmixing accuracy often leads to higher computational complexity. To solve this problem, we propose a novel double regression-based sparse unmixing model (DRSUM), which can obtain better unmixing results with lower computational complexity. DRSUM decomposes the complex objective function into two simple formulas and completes the unmixing process through two sparse regressions. The unmixing result of the first sparse regression is added as a constraint to the second. DRSUM is an open model, and we can add different constraints to improve the unmixing accuracy. In addition, we can perform appropriate preprocessing to further improve the unmixing results. Under this model, a specific algorithm called double regression-based sparse unmixing via K -means ( DRSU M K − means ) is proposed. The improved K -means clustering algorithm is first used for preprocessing, and then we impose single sparsity and joint sparsity (using l 2 , 0 norm to control the sparsity) constraints on the first and second sparse unmixing, respectively. To meet the sparsity requirement, we introduce the row-hard-threshold function to solve the l 2 , 0 norm directly. Then, DRSU M K − means can be efficiently solved under alternating direction method of multipliers (ADMM) framework. Simulated and real data experiments have proven the effectiveness of DRSU M K − means .

Joint sparsity-biased variational graph autoencoders

To bring the full benefits of machine learning to defense modeling and simulation, it is essential to first learn useful representations for sparse graphs consisting of both key entities (vertices) and their relationships (edges). Here, we present a new model, the Joint Sparsity-Biased Variational Graph AutoEncoder (JSBVGAE), capable of learning embedded representations of nodes from which both sparse network topologies and node features can be jointly and accurately reconstructed. We show that our model outperforms the previous state of the art on standard link-prediction and node-classification tasks, and achieves significantly higher whole-network reconstruction accuracy, while reducing the number of trained parameters.

Investigation of a multiphysics joint inversion framework

&lt;p&gt;We present a 3-D multiphysics joint inversion framework that in a certain sense is a trade-off between &amp;#8220;simultaneous&amp;#8221; and &amp;#8220;cooperative&amp;#8221; approaches to data integration. Using a variable splitting approach, inverse problems are solved by individual inverse solvers, on individual grids, while coupling combined with interpolation is implemented separately. Up to date, first-arrival seismic tomography, gravity and magnetotelluric inverse problems are included in this framework. Magnetotelluric inversion uses the NLCG method implemented in the ModEM code (Kelbert et al., 2014). Seismic tomography is based on the Gauss-Newton method with a finite-difference eikonal solver and posterior ray tracing. Gravity inversion uses the conjugate gradient method with wavelet compression of the sensitivity matrix. Structure coupling is based on mixed-norm regularization inducing joint sparsity between the models. Among particular functionals that were studied, following numerical experiments, joint total variation and joint minimum support have proved to&lt;span&gt; &lt;/span&gt;be the most efficient options. In the numerical experiments, we invert synthetics simulating regional datasets observed in Ireland.&lt;/p&gt;