Collaborative patch-based super-resolution for diffusion-weighted images

Abstract

To cite this version:Pierrick Coupé, José V Manjón, Maxime Chamberland, Maxime Descoteaux, Bassem Hiba. Collaborative patch-based super-resolution for diffusion-weighted images.. NeuroImage, Elsevier, 2013 AbstractIn this paper, a new single image acquisition super-resolution method is proposed to increase image resolution of diffusion weighted (DW) images. Based on a nonlocal patch-based strategy, the proposed method uses a non diffusion image (b0) t o c o n s t r a i n t h e r e c o n s t r u c t i o n o f DW i m a g e s .An extensive validation is presented with a gold standard built on averaging 10 high-resolution DW acquisitions. A comparison with classical interpolation methods such as trilinear and B-spline demonstrates the competitive results of our proposed approach in terms o f improvements on image reconstruction, fractional anisotropy (FA) estimation, generalized FA and angular reconstruction for tensor and high angular resolution diffusion imaging (HARDI) models. Besides, first results of reconstructed ultra high resolution DW images a r e p r e s e n t e d a t 0 . 6 x 0 . 6 x 0 . 6 m m 3 a n d 0.4x0.4x0.4mm 3 using our gold standard b a s e d o n the average of 10 acquisitions, and on a single acquisition. Finally, fiber tracking results show the potential of the proposed super-resolution approach t o a c c u r a t e l y analyze white matter brain architecture.

Bayesian Image Quality Transfer

“…Numerous machine-learning based methods have been proposed. For instance, [2,3] use example patches from HR images to super-resolve scalar MR and DW images respectively, with an explicitly defined generative model relating a HR patch to a LR patch and carefully crafted regularisation. Another generative approach is the sparse-representation methods [4,5], which construct a coupled library of HR and LR images from training data and solve the SR problem through projection onto it.…”

“…We incorporate Bayesian inference into the framework and name the new method Bayesian IQT (BIQT). Although many SR methods [2][3][4][5] can be cast as maximum a posteriori (MAP) optimisation problems, the dimensionality or complexity of the posterior distribution make the computation of uncertainty very expensive. In contrast, the random forest implementation of the original IQT is amenable to uncertainty estimation thanks to the simple linear model at each leaf node, but the current approach computes maximum likelihood (ML) solution.…”

Dictionary Based Super-Resolution for Diffusion MRI

“…We then calculated Pearson's correlation between intra-subject SC and FC to quantify the consistency between the two connectivity estimates. Using this correlation measure, we compared the proposed super-resolution approach with trilinear and spline interpolation in addition to an alternative super-resolution method; collaborative and locally adaptive super-resolution (CLASR) [9]. To the best of our knowledge, CLASR is the only existing single image super-resolution method for dMRI which is independent of the diffusion model employed.…”

“…Even though this method eliminates the need for multiple acquisitions, it is only geared towards estimating diffusion tensors, and cannot be easily extended to higher order diffusion models such as orientation distribution functions (ODFs). To the best of our knowledge, the only previous work that tackled the problem of super-resolving dMRI data from a single acquisition independent of the diffusion model was by Coupé et al [9]. Specifically, the authors showed that super-resolving b=0 (non-diffusion-weighted) image using a locally adaptive patch-based strategy, and using this high-resolution b=0 image to drive the reconstruction of DW images outperforms upsampling of dMRI data using classical interpolation methods.…”

A Hybrid and Adaptive Non-Local Means Wavelet based MRI Denoising Method with Bilateral Filter Enhancement

“…(24) is estimated as the restored value of square noisy image and 2 2  is a fixed signal independent bias which can be removed by subtracting it from each pixel in the magnitude squared image. (25) …”