Multiple Sparse Representations Classification.

Sparse representations classification (SRC) is a powerful technique for pixelwise classification of images and it is increasingly being used for a wide variety of image analysis tasks. The method uses sparse representation and learned redundant dictionaries to classify image pixels. In this empirical study we propose to further leverage the redundancy of the learned dictionaries to achieve a more accurate classifier. In conventional SRC, each image pixel is associated with a small patch surrounding it. Using these patches, a dictionary is trained for each class in a supervised fashion. Commonly, redundant/overcomplete dictionaries are trained and image patches are sparsely represented by a linear combination of only a few of the dictionary elements. Given a set of trained dictionaries, a new patch is sparse coded using each of them, and subsequently assigned to the class whose dictionary yields the minimum residual energy. We propose a generalization of this scheme. The method, which we call multiple sparse representations classification (mSRC), is based on the observation that an overcomplete, class specific dictionary is capable of generating multiple accurate and independent estimates of a patch belonging to the class. So instead of finding a single sparse representation of a patch for each dictionary, we find multiple, and the corresponding residual energies provides an enhanced statistic which is used to improve classification. We demonstrate the efficacy of mSRC for three example applications: pixelwise classification of texture images, lumen segmentation in carotid artery magnetic resonance imaging (MRI), and bifurcation point detection in carotid artery MRI. We compare our method with conventional SRC, K-nearest neighbor, and support vector machine classifiers. The results show that mSRC outperforms SRC and the other reference methods. In addition, we present an extensive evaluation of the effect of the main mSRC parameters: patch size, dictionary size, and sparsity level.

Interactive local super-resolution reconstruction of whole-body MRI mouse data: a pilot study with applications to bone and kidney metastases.

In small animal imaging studies, when the locations of the micro-structures of interest are unknown a priori, there is a simultaneous need for full-body coverage and high resolution. In MRI, additional requirements to image contrast and acquisition time will often make it impossible to acquire such images directly. Recently, a resolution enhancing post-processing technique called super-resolution reconstruction (SRR) has been demonstrated to improve visualization and localization of micro-structures in small animal MRI by combining multiple low-resolution acquisitions. However, when the field-of-view is large relative to the desired voxel size, solving the SRR problem becomes very expensive, in terms of both memory requirements and computation time. In this paper we introduce a novel local approach to SRR that aims to overcome the computational problems and allow researchers to efficiently explore both global and local characteristics in whole-body small animal MRI. The method integrates state-of-the-art image processing techniques from the areas of articulated atlas-based segmentation, planar reformation, and SRR. A proof-of-concept is provided with two case studies involving CT, BLI, and MRI data of bone and kidney tumors in a mouse model. We show that local SRR-MRI is a computationally efficient complementary imaging modality for the precise characterization of tumor metastases, and that the method provides a feasible high-resolution alternative to conventional MRI.

Three-dimensional inversion recovery manganese-enhanced MRI of mouse brain using super-resolution reconstruction to visualize nuclei involved in higher brain function.

The visualization of activity in mouse brain using inversion recovery spin echo (IR-SE) manganese-enhanced MRI (MEMRI) provides unique contrast, but suffers from poor resolution in the slice-encoding direction. Super-resolution reconstruction (SRR) is a resolution-enhancing post-processing technique in which multiple low-resolution slice stacks are combined into a single volume of high isotropic resolution using computational methods. In this study, we investigated, first, whether SRR can improve the three-dimensional resolution of IR-SE MEMRI in the slice selection direction, whilst maintaining or improving the contrast-to-noise ratio of the two-dimensional slice stacks. Second, the contrast-to-noise ratio of SRR IR-SE MEMRI was compared with a conventional three-dimensional gradient echo (GE) acquisition. Quantitative experiments were performed on a phantom containing compartments of various manganese concentrations. The results showed that, with comparable scan times, the signal-to-noise ratio of three-dimensional GE acquisition is higher than that of SRR IR-SE MEMRI. However, the contrast-to-noise ratio between different compartments can be superior with SRR IR-SE MEMRI, depending on the chosen inversion time. In vivo experiments were performed in mice receiving manganese using an implanted osmotic pump. The results showed that SRR works well as a resolution-enhancing technique in IR-SE MEMRI experiments. In addition, the SRR image also shows a number of brain structures that are more clearly discernible from the surrounding tissues than in three-dimensional GE acquisition, including a number of nuclei with specific higher brain functions, such as memory, stress, anxiety and reward behavior.

Super-resolution reconstruction using cross-scale self-similarity in multi-slice MRI.

In MRI, the relatively thick slices of multi-slice acquisitions often hamper visualization and analysis of the underlying anatomy. A group of post-processing techniques referred to as super-resolution reconstruction (SRR) have been developed to address this issue. In this study, we present a novel approach to SRR in MRI, which exploits the high-resolution content usually available in the 2D slices of MRI slice stacks to reconstruct isotropic high-resolution 3D images. Relying on the assumption of local self-similarity of anatomical structures, the method can be applied both to a single slice stack and to the combination of multiple slice stacks that differ in the orientation of their field of view. We evaluate the method quantitatively on synthetic brain MRI and qualitatively on MRI of the lungs. The results show that the method outperforms state-of-the-art MRI super-resolution methods.

Super-resolution methods in MRI: can they improve the trade-off between resolution, signal-to-noise ratio, and acquisition time?

Improving the resolution in magnetic resonance imaging comes at the cost of either lower signal-to-noise ratio, longer acquisition time or both. This study investigates whether so-called super-resolution reconstruction methods can increase the resolution in the slice selection direction and, as such, are a viable alternative to direct high-resolution acquisition in terms of the signal-to-noise ratio and acquisition time trade-offs. The performance of six super-resolution reconstruction methods and direct high-resolution acquisitions was compared with respect to these trade-offs. The methods are based on iterative back-projection, algebraic reconstruction, and regularized least squares. The algorithms were applied to low-resolution data sets within which the images were rotated relative to each other. Quantitative experiments involved a computational phantom and a physical phantom containing structures of known dimensions. To visually validate the quantitative evaluations, qualitative experiments were performed, in which images of three different subjects (a phantom, an ex vivo rat knee, and a postmortem mouse) were acquired with different magnetic resonance imaging scanners. The results show that super-resolution reconstruction can indeed improve the resolution, signal-to-noise ratio and acquisition time trade-offs compared with direct high-resolution acquisition.