Super-resolution methods have recently became popular due to their ability to generate isotropic high resolution images from multiple low resolution acquisitions. In this work, we developed and evaluated a convex programming solution to the super-resolution reconstruction and applied it to combine shifted thick slice T2 images into images with isotropic resolution. With this formulation, using phantom and volunteer experiments, we show that, it is possible to generate high resolution images with better resolution and accuracy compared to the previously developed methods.
Purpose
Super-resolution methods aim at reconstructing high-resolution (HR) information from coarse-scale measurements. These methods were introduced in the MR image reconstruction field in early 2000s1 and recently became popular in the fields of fetal MRI2,3 and diffusion-weighted MRI4, where low resolution (LR) images acquired with different orientations are combined to generate an isotropic resolution image. The forward problem of super-resolution can be represented by equation-1, where matrix A defines an operator that generates low-resolution images from the desired band-limited high-resolution image x. Traditionally, inspired by the super-resolution literature in image processing5, this equation is solved by what can be described as a shift-and-add (SAA)6 approach where the low resolution images are interpolated onto an HR grid (after motion correction) followed by restoration for blur and noise removal. Recently Candes et.al7 proposed casting the super-resolution solution as a simple convex problem (Equation-2) and showed that this solution gives the exact object location and amplitudes when the objects are seperated by 1/(2*fc+1) where fc is the maximum frequency of the low resolution measurements. They also showed that when the measurements are noisy, this solution results in a bounded extrapolation error that scales quadratically with the super-resolution factor. In this work, we assessed whether this convex algorithm can be used to increase the resolution of MR acquisition in the slice direction, via repeated measurements of shifted low-resolution acquisitions.
Equation 1: $$y_k=Ax_k+v_k$$
Equation
2: $$min_x
||X||_{TV} \quad \textrm{subject
to} \quad ||Ax-y||_1
<
δ$$
1.Greenspan, H., Oz, G., Kiryati, N. and Peled, S., 2002. MRI inter-slice reconstruction using super-resolution. Magnetic resonance imaging, 20(5), pp.437-446.
2. Kuklisova-Murgasova, M., Quaghebeur, G., Rutherford, M.A., Hajnal, J.V. and Schnabel, J.A., 2012. Reconstruction of fetal brain MRI with intensity matching and complete outlier removal. Medical image analysis, 16(8), pp.1550-1564.
3. Gholipour, A., Estroff, J.A. and Warfield, S.K., 2010. Robust super-resolution volume reconstruction from slice acquisitions: application to fetal brain MRI. IEEE transactions on medical imaging, 29(10), pp.1739-1758.
4. Scherrer, B., Gholipour, A. and Warfield, S.K., 2012. Super-resolution reconstruction to increase the spatial resolution of diffusion weighted images from orthogonal anisotropic acquisitions. Medical image analysis, 16(7), pp.1465-1476.
5. Park, S.C., Park, M.K. and Kang, M.G., 2003. Super-resolution image reconstruction: a technical overview. IEEE signal processing magazine, 20(3), pp.21-36.
6. Farsiu, S., Robinson, D., Elad, M. and Milanfar, P., 2003, November. Robust shift and add approach to superresolution. In Optical Science and Technology, SPIE's 48th Annual Meeting (pp. 121-130). International Society for Optics and Photonics.
7. Candès, E.J. and Fernandez-Granda, C., 2013. Super-resolution from noisy data. Journal of Fourier Analysis and Applications, 19(6), pp.1229-1254.
8. Michael Grant and Stephen Boyd. CVX: Matlab software for disciplined convex programming, version 2.0 beta. http://cvxr.com/cvx, September 2013.